WO1999021098A2 - Determining cluster membership in a distributed computer system - Google Patents

Determining cluster membership in a distributed computer system Download PDF

Info

Publication number
WO1999021098A2
WO1999021098A2 PCT/US1998/022161 US9822161W WO9921098A2 WO 1999021098 A2 WO1999021098 A2 WO 1999021098A2 US 9822161 W US9822161 W US 9822161W WO 9921098 A2 WO9921098 A2 WO 9921098A2
Authority
WO
WIPO (PCT)
Prior art keywords
nodes
node
proposed
computer system
proposed membership
Prior art date
Application number
PCT/US1998/022161
Other languages
French (fr)
Other versions
WO1999021098A3 (en
Inventor
Hossein Moiin
Ronald Widyono
Ramin Modiri
Original Assignee
Sun Microsystems, Inc.
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Sun Microsystems, Inc. filed Critical Sun Microsystems, Inc.
Priority to CA002306718A priority Critical patent/CA2306718A1/en
Priority to EP98953773A priority patent/EP1025506A2/en
Priority to JP2000517348A priority patent/JP2001521222A/en
Priority to AU11054/99A priority patent/AU1105499A/en
Publication of WO1999021098A2 publication Critical patent/WO1999021098A2/en
Publication of WO1999021098A3 publication Critical patent/WO1999021098A3/en

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F9/00—Arrangements for program control, e.g. control units
    • G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
    • G06F9/46—Multiprogramming arrangements
    • G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
    • G06F9/5061—Partitioning or combining of resources
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F11/00—Error detection; Error correction; Monitoring
    • G06F11/07—Responding to the occurrence of a fault, e.g. fault tolerance
    • G06F11/14—Error detection or correction of the data by redundancy in operation
    • G06F11/1402—Saving, restoring, recovering or retrying
    • G06F11/1415—Saving, restoring, recovering or retrying at system level
    • G06F11/142—Reconfiguring to eliminate the error
    • G06F11/1425—Reconfiguring to eliminate the error by reconfiguration of node membership
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
    • H04L41/08—Configuration management of networks or network elements
    • H04L41/0803—Configuration setting
    • H04L41/0823—Configuration setting characterised by the purposes of a change of settings, e.g. optimising configuration for enhancing reliability
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
    • H04L41/08—Configuration management of networks or network elements
    • H04L41/0866—Checking the configuration
    • H04L41/0873—Checking configuration conflicts between network elements
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
    • H04L41/08—Configuration management of networks or network elements
    • H04L41/0893—Assignment of logical groups to network elements
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L43/00—Arrangements for monitoring or testing data switching networks
    • H04L43/08—Monitoring or testing based on specific metrics, e.g. QoS, energy consumption or environmental parameters
    • H04L43/0805—Monitoring or testing based on specific metrics, e.g. QoS, energy consumption or environmental parameters by checking availability
    • H04L43/0811—Monitoring or testing based on specific metrics, e.g. QoS, energy consumption or environmental parameters by checking availability by checking connectivity
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F11/00—Error detection; Error correction; Monitoring
    • G06F11/07—Responding to the occurrence of a fault, e.g. fault tolerance
    • G06F11/16—Error detection or correction of the data by redundancy in hardware
    • G06F11/20—Error detection or correction of the data by redundancy in hardware using active fault-masking, e.g. by switching out faulty elements or by switching in spare elements
    • G06F11/202—Error detection or correction of the data by redundancy in hardware using active fault-masking, e.g. by switching out faulty elements or by switching in spare elements where processing functionality is redundant
    • G06F11/2023—Failover techniques
    • G06F11/2033—Failover techniques switching over of hardware resources
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F11/00—Error detection; Error correction; Monitoring
    • G06F11/07—Responding to the occurrence of a fault, e.g. fault tolerance
    • G06F11/16—Error detection or correction of the data by redundancy in hardware
    • G06F11/20—Error detection or correction of the data by redundancy in hardware using active fault-masking, e.g. by switching out faulty elements or by switching in spare elements
    • G06F11/202—Error detection or correction of the data by redundancy in hardware using active fault-masking, e.g. by switching out faulty elements or by switching in spare elements where processing functionality is redundant
    • G06F11/2038—Error detection or correction of the data by redundancy in hardware using active fault-masking, e.g. by switching out faulty elements or by switching in spare elements where processing functionality is redundant with a single idle spare processing component
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F11/00—Error detection; Error correction; Monitoring
    • G06F11/07—Responding to the occurrence of a fault, e.g. fault tolerance
    • G06F11/16—Error detection or correction of the data by redundancy in hardware
    • G06F11/20—Error detection or correction of the data by redundancy in hardware using active fault-masking, e.g. by switching out faulty elements or by switching in spare elements
    • G06F11/202—Error detection or correction of the data by redundancy in hardware using active fault-masking, e.g. by switching out faulty elements or by switching in spare elements where processing functionality is redundant
    • G06F11/2046—Error detection or correction of the data by redundancy in hardware using active fault-masking, e.g. by switching out faulty elements or by switching in spare elements where processing functionality is redundant where the redundant components share persistent storage
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F2209/00—Indexing scheme relating to G06F9/00
    • G06F2209/50—Indexing scheme relating to G06F9/50
    • G06F2209/505—Clust
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04L—TRANSMISSION OF DIGITAL INFORMATION, e.g. TELEGRAPHIC COMMUNICATION
    • H04L41/00—Arrangements for maintenance, administration or management of data switching networks, e.g. of packet switching networks
    • H04L41/06—Management of faults, events, alarms or notifications

Definitions

  • the present invention relates to fault tolerance in distributed computer systems and, in particular, to a particularly robust mechanism for determining which nodes in a failing distributed computer system form a cluster and have access to shared resources.
  • the problems associated with providing membership services in a distributed computer system have generated a considerable amount of interest in both academic and industrial fronts.
  • the Parallel Database (PDB) system available from Sun Microsystems, Inc. of Palo Alto, California, being a distributed system, has used the cluster membership monitor to provide mechanisms to keep track of the member nodes and to coordinate the reconfiguration of the cluster applications and services when the cluster membership changes.
  • PDB Parallel Database
  • a further impact of the configuration of external devices is the issue of failure fencing.
  • the shared resources (often disks) are fenced against intervention from nodes that are not part of the cluster.
  • the issue of fencing was simple due to the fact that only two nodes existed in a cluster and they were connected to all the shared resources. The node that remained in the cluster would reserve all the shared resources and would disallow the non-member node from accessing these resources until that node became part of the cluster. Such a simple operation is not possible for an architecture in which all disks are not connected to all nodes.
  • CMM Cluster Membership Monitor
  • a node can join a cluster after the node is restarted and after other members of the cluster accepted it as a new member, following a reconfiguration.
  • the cluster membership monitor handles communication failures that isolate one or more nodes from those nodes with a majority quorum. Note that the detection of the communication failure, i.e. detecting that the communication graph is not fully connected, is the responsibility of the communication monitor which is not part of the membership monitor. It is assumed that the communication monitor will notify the membership monitor of communication failures and that the membership monitor will handle this via a reconfiguration.
  • the CMM does not guarantee the health of the overall system or that the applications are present on any given node.
  • the only guarantees made by the CMM is that the system's hardware is up and running and that the operating system is present and functioning.
  • failures are considered in the design of the system. There are three failures that we consider; node failures, communication failures, and device failures. Note that the failures of the client nodes, terminal concentrators, and the administration workstation are not considered to be failures within "our" system.
  • Node Failures A node is considered to have failed when it stops sending its periodic heart-beat messages (SCI or CMM) to other members of the duster. Furthermore, nodes are considered to behave in a non-malicious fashion, a node that is considered failed by the system will not try to intentionally send conflicting information to other members of the cluster. It is possible for nodes to fail intermittently, as in the case of a temporary dead-lock, or to be viewed as failed by only part of the remaining system, as in the case of a failed adaptor or switch. The cluster membership monitor should be able to handle all these cases and should remove failed nodes from the system within a bounded time.
  • SCI or CMM periodic heart-beat messages
  • the private communication medium may fail due to a failure of a switch, a failure of an adaptor card, a failure of the cable, o failure of various software layers. These failures are masked by the cluster communication monitor (CCM or CIS) so that the cluster membership monitor does not have to deal with the specific failure. In addition, the cluster membership monitor will either send its messages through all available links of the medium. Hence, failure of any individual link does not affect the correct operation of the CMM and the only communication failure affecting CMM is the total loss of communication with a member node. This is equivalent to a node failure as there are no physical paths to send a heart-beat message over the private communication medium.
  • CCM cluster communication monitor
  • n is the number of nodes in the system.
  • Device Failures Devices that affect the operation of the cluster membership monitor are the quorum devices. Traditionally these have been disk controllers on the Sparc Storage Arrays (SSA's), however, in some distributed computer systems, a disk can also be used as a quorum device. Note that the failure of the quorum device is equivalent to the failure of a node and that the CMM in some conventional systems will not use a quorum device unless it is running on a two node cluster.
  • SSA Sparc Storage Arrays
  • Some distributed computer systems are specified to have no single point of failures. Therefore, the system must tolerate the failure of a single node as well as consecutive failures of n - 1 nodes of the system. Given the above discussion on communication failures, this specification implies that we cannot tolerate the total loss of the communication medium in such a system. While it may not be possible, or desirable, to tolerate a total loss of the private communication medium, it should be possible to tolerate more than a single failure at any given time. First, let us define what a cluster is and how various failures affect it.
  • a cluster is defined as having N nodes, a private communication medium, and a quorum mechanism, where the total failure of the private communication medium is equivalent to the failure of N - 1 nodes and the failure of the quorum mechanism is equivalent to the failure of one node.
  • a cluster with N nodes, where N > 3, a private communication medium, and a quorum mechanism, should be able to provide services and access to the data, however partial, in the case of ⁇ f /2 - 1 node failures.
  • the cluster can tolerate only one of the following failures:
  • cluster membership in a distributed computer system is determined by determining with which other nodes each node is in communication and distributing that connectivity information through the nodes of the system. Accordingly, each node can determine an optimized new cluster based upon the connectivity information. Specifically, each node has information regarding with which nodes the node is in communication and similar information for each other node of the system. Therefore, each node has complete information regarding interconnectivity of all nodes which are directly or indirectly connected.
  • Each node applies optimization criteria to such connectivity information to determine an optimal new cluster.
  • Data represent the optimal new cluster is broadcast by each node.
  • the optimal new cluster determined by the various nodes are collected by each node.
  • each node has data representing the proposed new cluster which is perceived by each respective node to be optimal.
  • Each node uses such data to elect a new cluster from the various proposed new clusters. For example, the new cluster represented by more proposed new clusters than any other is elected as the new cluster. Since each node receives the same proposed new clusters from the potential member nodes of the new cluster, the new cluster membership is reached unanimously. In addition, since each node has more complete information regarding the potential member nodes of the new cluster, the resulting new cluster consistently has a relatively optimal configuration.
  • Figure 1 is a block diagram of a distributed computer system in which communications between two nodes and two respective switches have failed.
  • Figure 2 is a block diagram of a distributed computer system which includes dual- ported devices.
  • the cluster membership can be partitioned into two or more fully-connected subsets of nodes having a majority of the votes, a minority of the votes, or exactly half of the votes.
  • the first two cases may be resolved by only allowing a subset having a majority vote to form the next generation of the cluster. In the latter case, a tie breaking mechanism must be employed.
  • Some cluster membership algorithms take advantage of the restrictions imposed by a two node architecture in resolving these issues. To generalize for architectures involving more than two nodes, the following new issues are resolved by the algorithm according to the present invention.
  • this external device is a disk or a controller that resides on the SSA's.
  • the choice of a quorum device, particularly for a disk, has some unfavorable properties which adversely affect the overall availability of the cluster.
  • the original algorithm considers the explicit simultaneous cluster shutdown of more than half the nodes to be equivalent to a partition excluding those nodes. To avoid the resulting loss of quorum and the complete cluster shutdown, the new algorithm uses the notification of an explicit shutdown by a node to reduce the quorum requirement.
  • the original algorithm does not reach an agreement.
  • the algorithm would converge when a set of nodes agree on the same membership proposal, but this condition is never satisfied.
  • this condition is suspected (using a timeout)
  • some nodes modify their membership proposal to a subset that is maximally connected.
  • the membership monitors on different nodes of a cluster exchange messages with each other to notify that they are alive, i.e. exchange heart-beats, and to initiate a cluster reconfiguration. While it is possible to distinguish between these two types of messages, in practice they are the same message and we refer to them as RECONF msg messages to stress that they cause reconfiguration on the receiving nodes.
  • V t A vector that contains the connectivity information of node / ' .
  • SD A vector that contains node t's view of nodes that have voluntarily left the cluster as of the time the most recent stable membership was established.
  • the membership algorithm assumes that the cluster is made up of equally valuable nodes, i.e. that the cluster is a homogeneous cluster.
  • the membership algorithm is based on a set of rules that are stated in the following precedence order and are used to develop the membership algorithm:
  • a node must include itself in its proposed set.
  • a node will vote for nodes that are already in the cluster over the ones that are trying to join it.
  • a node will propose a set that includes itself and has the maximum number of fully connected nodes.
  • All nodes agree on a statically defined preference order among nodes, e.g. lower numbered nodes are preferred to higher numbered ones.
  • the above set of rules define a hierarchy of rules with the statically defined preference being at the bottom of such a hierarchy. Note that at the above set of rules also defines an optimal membership set, A.
  • Joins This is when nodes either form a new cluster or join an already existing one.
  • First Join This is done only for the first node of a cluster and is implemented via a new command, pdbadmin startcluster, which signals to the CMM running at that node that it should not expect to hear from other members of the cluster, as there are not any. This command can only be issued once in the beginning of the life of a cluster, where the life of a cluster is defined as the time that spans from the moment that the pdbadmin startcluster is issued to the time where the cluster returns to having no members.
  • Leaves This is when nodes that were members of the cluster leave the cluster either voluntarily or involuntarily.
  • (a) Voluntarily Leaves The operator issues a pdbadmin stopnode command to a node, or a set of nodes. This will cause the affected nodes to go through the stop sequence which would result in the node sending a message to all other nodes in the cluster informing of them that it is leaving the cluster. This information can be used, and is used, by the membership algorithm to optimize certain aspects of the membership.
  • the node can send a message, the same one as the voluntarily leaves, that would inform other members of the cluster that this node will not be part of the cluster.
  • the same optimizations that can be performed for the voluntarily leaving of a node can actually be implemented here.
  • a node may leave the cluster due to a request from an application program with the appropriate privileges, ⁇ .
  • the node does not complete its abort sequence and panics the system. This is the most difficult of all failures to deal with and is usually detected by the absence of a heart-beat message from the failed node. This failure is un-distinguishable from a network failure in an asynchronous distributed system.
  • Each node, / ' will update its connectivity state matrix, C Being as soon as it hears from a node.
  • the matrix C is node 's understanding of the overall connectivity of the system. If node / does not hear from a nodey within the specified time, or is informed that node is down or unreachable, it will mark the e,, element of C, as zero. Furthermore, it will mark all the elements of theyth row as NULL, which implies that node / does not have any information about the connectivity of node /. For other rows of the matrix, node / will update them by replacing the kth row of its connectivity matrix with the connectivity vector V*- that it receives from node k.
  • Each node / ' will initially include its /th row of C, in its RECONF msg as the
  • M prop proposed by node / ' is different from the vector V,.
  • ' is a proposed set that states a node's vote for other nodes in a binary form, whereas V, is a state vector
  • MP P which deals with the connectivity of the nodes in the system. Note that ' will have different elements, each element being a node id and a binary vote value, than V View when nodes cannot agree on a stable membership and a subset of V, needs to be proposed as the new membership set.
  • Each node keeps the total number of the nodes that are in its current view of the cluster membership, whether agreed or proposed, in a local variable N,. Note that N, is subject to the following rules during the execution of the membership algorithm:
  • M prop (a)N is initialized to the cardinality of .
  • N is incremented for each node that is trying to join the cluster. (One increment per node as enforced via the nodeid check which is embedded in the message, done by the receiver thread.)
  • (c)N is decremented for each node that aborts, as defined in part 2(b)i of Section 4.3, or voluntarily leaves. (Done by the receiver thread.)
  • Each node, i, prior to entering the membership algorithm will have a sequence number, seq num, that is the same for all nodes in the current cluster.
  • each node will have its connectivity state matrix, C,. Note that C, is a n x n matrix, where n is the maximum number of nodes as defined by the current cluster configuration file, i.e. the current cdb file.
  • Each node that is a joiner will have a variable, joining_node, set to TRUE.
  • M agreed node has become a member of ' , it is no longer a joiner, and joining ⁇ node is set to FALSE.
  • a node that is executing the first join will have a variable, start_cluster, set to TRUE. Nodes that are trying to join the cluster will have their start cluster variable initialized to FALSE.
  • a joiner may not join a partially connected or nonexistent */
  • joining node FALSE
  • the function membership_proposal() returns a membership proposal based on C / , including all nodes that are not in the DOWN state. It also excludes
  • M prop set ' is agreed upon by all the members of that set. In order to count the number of
  • the fijnction share_quorum_dev() is implemented by using the CCD dynamic file and informs the membership algorithm of the cases, such as a two node cluster, in which two nodes do share a quorum device.
  • the binary function reserve_quorumQ returns false if and only if the device is already reserved by another node.
  • the function ait_for_user_input() is discussed in greater detail below.
  • the message would identify, for the appropriate nodes, the sets X or Y that must be shut down or informed to stay up.
  • the operator will break the tie by issuing the command pdbadmin stopnode to one set of nodes while issuing the new command pdbadmin continue to the other set.
  • the set that receives the stop command will abort, while the other set will stop printing the messages and continue its reconfiguration.
  • the operator can issue a clustm reconfigure command, which is a valid option if there was a communication break down and the operator has fixed it. Issuing a clustm reconfigure command will cause a new reconfiguration to take place.
  • the command reader thread will not signal the transitions thread that is waiting for one of those commands and will simply ignore the command.
  • the print thread meanwhile, will be printing these messages continuously once every few seconds, to inform the operator that some immediate action is required.
  • Node receives a message that indicates that a remote node in its current membership set has gone down.
  • Node has not received a message from a remote node in its current membership set for node down timeout.
  • failure fencing mechanism Another component of the system that requires modification due to the new architecture is the failure fencing mechanism that is employed in some distributed computer systems.
  • failure fencing mechanism that is employed in some distributed computer systems.
  • the solution provided is generic in the sense that it handles the various array topologies-cascaded, n + 1, cross-connected, and others-as well as different software configurations-CVM with Netdisk, stand alone VxVM, or others. It also handles the 2-node cross connected array configuration without treating it as a special case.
  • every node has a backup node.
  • the backup node becomes the master of the set of NetDisk devices owned by a failed node.
  • the backup node becomes the primary owner of the set of disk group resources owned by a failed node.
  • N denote a node of a cluster
  • D denote a set of storage devices (composed of one or more SSA and/or Multipacks).
  • CCD maintains this information in its database and can be queried to obtain this information and. also whether the currently reconfiguring node is the backup of the failed node.
  • information about the primary ownership of a disk group is maintained in the cdb file in the following format:
  • the primary owner of cluster disk-groups dgl and dg2 is node 0 and its backup node is node 1.
  • node N owns resources whose primary owners are part of the cluster membership and have it undertake the appropriate actions. If the algorithms are implemented correctly, at no point in time should node N, own resources belonging to node N, if N, was already part of the cluster. This is a safe assumption required for correctness and integrity. This algorithm is slightly expensive in terms of reconfiguration times, but in no way it constitutes a bottleneck.
  • Resources that need to be highly available are migrated over to a surviving node from a failed one.
  • examples of such resources are disk groups in a shared nothing database environment, disk groups for HA- NFS file-systems and, logical IP addresses.
  • Arbitrary resources can be designated in the CCD with a master and a backup node.
  • the logical IP addresses can be migrated from failed nodes to surviving ones. Note that for a switch-over to take place, the backup node would have to release the resources of the joining node one step before the joining node takes over its resources.
  • the node N 2 has a disk group G whose devices are scattered on disks in arrays D and D3. lfN 2 now fails, G cannot be imported in its entirety on either Nj or N3 since all of its disks won't be visible on either N or N3. Such configurations are not supported in some distributed computer systems. If a node owns a disk group and if the node fails, it should be possible for the disk group in its entirety to be taken over by one of the surviving nodes. This does not constrain the array topology, but places restraints on how data is scattered across the arrays.

Abstract

Cluster membership in a distributed computer system is determined by determining with which other nodes each node is in communication and distributing that connectivity information through the nodes of the system. Accordingly, each node can determine an optimized new cluster based upon the connectivity information. Specifically, each node has information regarding with which nodes the node is in communication and similar information for each other node of the system. Therefore, each node has complete information regarding interconnectivity of all nodes which are directly or indirectly connected. Each node applies optimization criteria to such connectivity information to determine an optimal new cluster. Data represent the optimal new cluster is broadcast by each node. In addition, the optimal new cluster determined by the various nodes are collected by each node. Thus, each node has data representing the proposed new cluster which is perceived by each respective node to be optimal. Each node uses such data to elect a new cluster from the various proposed new clusters. For example, the new cluster represented by more proposed new clusters than any other is elected as the new cluster. Since each node receives the same proposed new clusters from the potential member nodes of the new cluster, the new cluster membership is reached unanimously. In addition, since each node has more complete information regarding the potential member nodes of the new cluster, the resulting new cluster consistently has a relatively optimal configuration.

Description

DETERMININGCLUSTER MEMBERSHIP IN ADISTRIBUTED COMPUTER
SYSTEM
SPECIFICATION
FIELD OF THE INVENTION
The present invention relates to fault tolerance in distributed computer systems and, in particular, to a particularly robust mechanism for determining which nodes in a failing distributed computer system form a cluster and have access to shared resources.
BACKGROUND OF THE INVENTION
The problems associated with providing membership services in a distributed computer system have generated a considerable amount of interest in both academic and industrial fronts. The Parallel Database (PDB) system available from Sun Microsystems, Inc. of Palo Alto, California, being a distributed system, has used the cluster membership monitor to provide mechanisms to keep track of the member nodes and to coordinate the reconfiguration of the cluster applications and services when the cluster membership changes. Herein, we define the general problem of membership in a cluster of computers where the nodes of the cluster may not be fully connected and we propose a solution to it.
The general problem of membership can be encapsulated by the design goals for the membership algorithm that are outlined below. We will further describe the problems that we are trying to address after we state these goals.
1. A uniform and robust membership algorithm regardless of the system architecture that is able to tolerate consecutive failures of nodes, links, storage devices or the communication medium. Stated in other words, no single point of failure should result in cluster unavailability.
2. Data integrity is never jeopardized even in the presence of multiple and simultaneous faults. This is accomplished by:
(a) Having only one cluster with majority quorum operational at any given time. (b) The cluster with majority quorum should never reach inconsistent agreement.
(c) Removal of isolated and faulty nodes from the cluster in a bounded time.
(d) Timely fencing of non-member nodes from the shared resources.
The hardware architecture of some conventional distributed computer systems poses specific problems for the membership algorithm. For example, consider the configuration shown in Figure 1. In this figure each of nodes 100A-D is supposed to be connected to two switches 101-102; however, there are two link failures that effectively disallow nodes 100 A and 100D from communicating with each other. Some conventional membership algorithms are not capable of dealing with such a failure and will not reach an agreement on a surviving majority quorum. Those algorithms assume that the nodes are fully connected and do not deal with the problem of a partitioned network. What is needed is a generalized algorithm that deals with the issue of a partitioned network as well as networks that are not partitioned.
Further complications arise when we need to make decisions about split-brain, or possible split-brain situations. For example, consider the configuration shown in Figure 2. In this configuration if the communication between nodes {200A, 200B} and {200C, 200D} is lost so that there are two sub-clusters with equal number of nodes, then the current quorum algorithm may lead to the possible shut-down of the entire cluster. Other situations during which the current algorithm is not capable of dealing with include when there are two nodes in the system and they do not share an external device.
The above examples illustrate a new set of problems for the membership and quorum algorithms that were not possible under the more simplistic architecture of some conventional distributed computer systems where a fully connected network was assumed. Our approach to solving these new problems is to integrate the membership and quorum algorithms more closely and to provide a flexible algorithm that would maximize the cluster availability and performance as viewed by the user.
A further impact of the configuration of external devices is the issue of failure fencing. In a clustered system the shared resources (often disks) are fenced against intervention from nodes that are not part of the cluster. In some distributed computer systems, the issue of fencing was simple due to the fact that only two nodes existed in a cluster and they were connected to all the shared resources. The node that remained in the cluster would reserve all the shared resources and would disallow the non-member node from accessing these resources until that node became part of the cluster. Such a simple operation is not possible for an architecture in which all disks are not connected to all nodes. Given that the SPARC Storage Arrays (SSA's) are only dual ported, there needs to be a new way that would effectively fence a non-member node out of the shared resources. The Cluster Membership Monitor, CMM, which is responsible for the membership, quorum and failure fencing algorithms, handles state transitions which lead to changes in the membership. These transitions are listed below.
• Failure of a Node: When a node fails, the remaining nodes will initiate a cluster reconfiguration resulting in a membership that will not include the failed node.
• Joining of a Node: A node can join a cluster after the node is restarted and after other members of the cluster accepted it as a new member, following a reconfiguration.
• Voluntary Leave: A node can leave the cluster voluntarily, and the remaining members of the cluster will reconfigure into the next generation of the cluster.
• Communication Failures: The cluster membership monitor handles communication failures that isolate one or more nodes from those nodes with a majority quorum. Note that the detection of the communication failure, i.e. detecting that the communication graph is not fully connected, is the responsibility of the communication monitor which is not part of the membership monitor. It is assumed that the communication monitor will notify the membership monitor of communication failures and that the membership monitor will handle this via a reconfiguration.
It is also important to note that the CMM does not guarantee the health of the overall system or that the applications are present on any given node. The only guarantees made by the CMM is that the system's hardware is up and running and that the operating system is present and functioning.
We would like to explicitly define what failures are considered in the design of the system. There are three failures that we consider; node failures, communication failures, and device failures. Note that the failures of the client nodes, terminal concentrators, and the administration workstation are not considered to be failures within "our" system.
Node Failures: A node is considered to have failed when it stops sending its periodic heart-beat messages (SCI or CMM) to other members of the duster. Furthermore, nodes are considered to behave in a non-malicious fashion, a node that is considered failed by the system will not try to intentionally send conflicting information to other members of the cluster. It is possible for nodes to fail intermittently, as in the case of a temporary dead-lock, or to be viewed as failed by only part of the remaining system, as in the case of a failed adaptor or switch. The cluster membership monitor should be able to handle all these cases and should remove failed nodes from the system within a bounded time. Communication Failures: The private communication medium may fail due to a failure of a switch, a failure of an adaptor card, a failure of the cable, o failure of various software layers. These failures are masked by the cluster communication monitor (CCM or CIS) so that the cluster membership monitor does not have to deal with the specific failure. In addition, the cluster membership monitor will either send its messages through all available links of the medium. Hence, failure of any individual link does not affect the correct operation of the CMM and the only communication failure affecting CMM is the total loss of communication with a member node. This is equivalent to a node failure as there are no physical paths to send a heart-beat message over the private communication medium. It is important to note that in a switched architecture, such as in the 2.0 release of Energizer, the failure of all switches is logically equivalent to the simultaneous failure of n - 1 nodes where n is the number of nodes in the system. Device Failures: Devices that affect the operation of the cluster membership monitor are the quorum devices. Traditionally these have been disk controllers on the Sparc Storage Arrays (SSA's), however, in some distributed computer systems, a disk can also be used as a quorum device. Note that the failure of the quorum device is equivalent to the failure of a node and that the CMM in some conventional systems will not use a quorum device unless it is running on a two node cluster.
Some distributed computer systems are specified to have no single point of failures. Therefore, the system must tolerate the failure of a single node as well as consecutive failures of n - 1 nodes of the system. Given the above discussion on communication failures, this specification implies that we cannot tolerate the total loss of the communication medium in such a system. While it may not be possible, or desirable, to tolerate a total loss of the private communication medium, it should be possible to tolerate more than a single failure at any given time. First, let us define what a cluster is and how various failures affect it.
A cluster is defined as having N nodes, a private communication medium, and a quorum mechanism, where the total failure of the private communication medium is equivalent to the failure of N - 1 nodes and the failure of the quorum mechanism is equivalent to the failure of one node.
Now we can make the following fault-tolerance goal for the cluster membership monitor;
A cluster with N nodes, where N > 3, a private communication medium, and a quorum mechanism, should be able to provide services and access to the data, however partial, in the case of ιf /2 - 1 node failures. For a two node cluster the cluster can tolerate only one of the following failures:
• Loss of one of its nodes.
• Loss of the private communication medium. Note that this case is logically equivalent to the loss of one node.
• Loss of the quorum device.
• Loss of one of its nodes and the communication medium. Note that this case is logically equivalent to the loss of one node.
Note that the total loss of the communication medium in a system with more than 2 nodes is indeed a double failure (as both switches would have to be non-operational) and the system is not required to tolerate such a failure.
SUMMARY OF THE INVENTION
In accordance with the present invention, cluster membership in a distributed computer system is determined by determining with which other nodes each node is in communication and distributing that connectivity information through the nodes of the system. Accordingly, each node can determine an optimized new cluster based upon the connectivity information. Specifically, each node has information regarding with which nodes the node is in communication and similar information for each other node of the system. Therefore, each node has complete information regarding interconnectivity of all nodes which are directly or indirectly connected.
Each node applies optimization criteria to such connectivity information to determine an optimal new cluster. Data represent the optimal new cluster is broadcast by each node. In addition, the optimal new cluster determined by the various nodes are collected by each node. Thus, each node has data representing the proposed new cluster which is perceived by each respective node to be optimal. Each node uses such data to elect a new cluster from the various proposed new clusters. For example, the new cluster represented by more proposed new clusters than any other is elected as the new cluster. Since each node receives the same proposed new clusters from the potential member nodes of the new cluster, the new cluster membership is reached unanimously. In addition, since each node has more complete information regarding the potential member nodes of the new cluster, the resulting new cluster consistently has a relatively optimal configuration.
BRIEF DESCRIPTION OF THE DRAWINGS
Figure 1 is a block diagram of a distributed computer system in which communications between two nodes and two respective switches have failed.
Figure 2 is a block diagram of a distributed computer system which includes dual- ported devices.
DETAILED DESCRIPTION
Agreement among the processors of a distributed system on which processors are members of that system is a fundamental problem in the design of highly-available distributed systems. Changes in membership may occur when processors are shut down, fail, recover, or new processors are added. Currently, there is no agreed definition of the processor membership problem and the existing membership protocols provide significantly different guarantees about their services. The protocols we are interested in lie toward the stricter end of the spectrum, ensuring that processor in the current membership agree on the set of member nodes and that membership changes occur at logically equivalent times in different nodes.
With the failures described above, the cluster membership can be partitioned into two or more fully-connected subsets of nodes having a majority of the votes, a minority of the votes, or exactly half of the votes. The first two cases may be resolved by only allowing a subset having a majority vote to form the next generation of the cluster. In the latter case, a tie breaking mechanism must be employed. Some cluster membership algorithms take advantage of the restrictions imposed by a two node architecture in resolving these issues. To generalize for architectures involving more than two nodes, the following new issues are resolved by the algorithm according to the present invention.
1. Resolving quorum and membership when not all pairs of nodes share a common external device.
Integration of quorum and membership algorithms is sometimes necessary for systems with more than two nodes and would result in some modifications in the membership algorithm. There is really no need for an external device to resolve membership and quorum issues if there are more than 2 nodes in a distributed system. However, a system with only two nodes needs an external quorum mechanism.
In some distributed computer systems, this external device is a disk or a controller that resides on the SSA's. The choice of a quorum device, particularly for a disk, has some unfavorable properties which adversely affect the overall availability of the cluster.
The situation for a 4 node system with an architecture that does not allow all nodes to be connected to all external devices can be more complicated. In such an architecture, some combination of nodes that can form a cluster do not share any external devices, other than the communication medium, and therefore if we are to allow for such clusters to exist, we would need an alternative quorum mechanism. Given that the use of the public network is a serious security hole, we are left with the last resort, i.e. the human intervention. We sometimes use this resort as described more completely below for cases in which a winner by the majority of votes cannot be automatically determined and explain a new user interface in greater detail below.
2. Allowing for a majority quorum requirement that adapts to changing memberships. With more than two configured nodes, requiring more than half the total configured votes for majority quorum would limit the flexibility for the user. In a four node system, not even two nodes could form a cluster. The modified algorithm bases the quorum requirement on the votes of the current membership and of any joining nodes.
3. Treating the "voluntary leave" of a cluster member as a hint that the majority quorum requirement may be lowered.
The original algorithm considers the explicit simultaneous cluster shutdown of more than half the nodes to be equivalent to a partition excluding those nodes. To avoid the resulting loss of quorum and the complete cluster shutdown, the new algorithm uses the notification of an explicit shutdown by a node to reduce the quorum requirement.
4. Dealing with joins when the nodes are partitioned.
With a two node configuration and a tie-breaking quorum device, it is not possible for the two nodes to form independent clusters when communication between them is broken. With more than two nodes and the dynamic quorum requirement modification in item 2, such an inconsistent state (two or more independent clusters) is possible, since all fully connected subsets of nodes can form a cluster with quorum. The algorithm according to the present invention differentiates between initial creation of the cluster and subsequent joins. Except for this initial join, joins cannot independently form a cluster-a node can only join an existing cluster. The user interface for this is discussed below.
5. Handling failures that occur during the membership algorithm.
With a dynamic quorum requirement, inconsistencies among nodes in the number of votes required for quorum can occur when failures happen during a reconfiguration. To avoid the possibility of two or more subsets having quorum and forming independent clusters, the modified algorithm imposes a restriction on joiners-joiners may only join a fijlly intact existing cluster. Note that this requires all nodes participating in the membership algorithm to agree on the membership of the existing cluster.
6. Dealing with the partial connectivity situation depicted in Figure 1.
In such a scenario, the original algorithm does not reach an agreement. The algorithm would converge when a set of nodes agree on the same membership proposal, but this condition is never satisfied. In the algorithm according to the present invention, when this condition is suspected (using a timeout), some nodes modify their membership proposal to a subset that is maximally connected.
In the following subsections, we discuss the format of the messages that cluster daemons exchange, define what an optimal membership set is and how to select one, specify the assumptions made in the membership algorithm in addition to those made above, describe how a change in membership may come about, describe the membership algorithm, explain how CMM can suspend and resume a set of registered processes, discuss how CMM checks for consistency of its configuration database, and specify some new user interfaces that may be needed.
4.1 CMM Messages
The membership monitors on different nodes of a cluster exchange messages with each other to notify that they are alive, i.e. exchange heart-beats, and to initiate a cluster reconfiguration. While it is possible to distinguish between these two types of messages, in practice they are the same message and we refer to them as RECONF msg messages to stress that they cause reconfiguration on the receiving nodes.
Each RECONF msg will include the following fields:
• A sequence number, seq nim, that distinguishes between different reconfigurations.
• A vector, Mit that contains node t's membership vote.
• A vector S, that contains node 's view of the most recent stable membership.
• A vector, Vt, that contains the connectivity information of node /'. • A vector SD, that contains node t's view of nodes that have voluntarily left the cluster as of the time the most recent stable membership was established.
• The state of the node St,.
• The node id of the originating node.
• A vector J„ that contains node t's view of nodes that are attempting to join.
• A flag indicating whether the originating node considers itself a joining node.
4.2 Definitions and Assumptions
The membership algorithm assumes that the cluster is made up of equally valuable nodes, i.e. that the cluster is a homogeneous cluster. The membership algorithm is based on a set of rules that are stated in the following precedence order and are used to develop the membership algorithm:
1. A node must include itself in its proposed set.
2. A node will vote for nodes that are already in the cluster over the ones that are trying to join it.
3. A node will propose a set that includes itself and has the maximum number of fully connected nodes.
4. All nodes agree on a statically defined preference order among nodes, e.g. lower numbered nodes are preferred to higher numbered ones.
The above set of rules define a hierarchy of rules with the statically defined preference being at the bottom of such a hierarchy. Note that at the above set of rules also defines an optimal membership set, A.
To find the optimal membership set, Λ, in a cluster with one or more failures is a computationally expensive task. This problem can be stated in terms of selecting the optimal, according to the definition of optimality derived from the above rules, subset of a set of nodes. Assuming that the cluster is composed of N nodes, finding Λ is equivalent to finding an optimal matrix of siz MxM from a matrix of size NxN, where M <N. This problem is indeed the well known problem of "N choose M" which is also known as binomial coefificients. The solution to this problem is ϊOI^) complexity if we assume that the system is homogeneous so that each node can be represented by either 0, 1 or - 1. While the cost of finding an optimal subset is prohibitively high for large N; for N < 20 this cost is not prohibitive. Therefore, for systems with 16 or less nodes, we propose to find the optimal set via an exhaustive search method. For systems with more than 20, nodes a heuristic algorithm that will approximate the optimal solution is preferred.
An assumption that we make is that when a node that aborts broadcasts a RECONF msg to all other members of the current cluster. We also assume that a node that wants to join the cluster does so in its begin state and its sequence number is reset to 0. We also assume that only messages with higher or equal sequence numbers to a node's own sequence number are processed and a state that is at most one behind in ordinal value. However, there is a significant exception. If the message comes from a node with the 'joiner' flag set, it will be processed even if the state is stale (more than one behind). These are the nodes that are trying to join the cluster and we must accept their initial messages. All these assumptions are enforced by our implementation of the membership algorithm according to the present invention.
4.3 Changes in the Membership
There are a number of ways in which a node can get into a reconfiguration which may result in a change in the membership according to the algorithm presented in the next subsection. The following is a list of them:
1. Joins: This is when nodes either form a new cluster or join an already existing one. (a) First Join: This is done only for the first node of a cluster and is implemented via a new command, pdbadmin startcluster, which signals to the CMM running at that node that it should not expect to hear from other members of the cluster, as there are not any. This command can only be issued once in the beginning of the life of a cluster, where the life of a cluster is defined as the time that spans from the moment that the pdbadmin startcluster is issued to the time where the cluster returns to having no members. If additional pdbadmin startcluster commands are issued they may cause, at worst the system to compromise data integrity if the node is isolated or, in more realistic cases, generate errors and abort the node for which this command was mistakenly issued, (b) Joins Following the First Join: These joins are done via the common pdbadmin startnode command and result in one node, or a set of nodes, to join the cluster. The nodes that are trying to join the cluster will communicate with the nodes that are already members of the cluster and try to see if they can join by going through the membership algorithm.
2. Leaves: This is when nodes that were members of the cluster leave the cluster either voluntarily or involuntarily.
(a) Voluntarily Leaves: The operator issues a pdbadmin stopnode command to a node, or a set of nodes. This will cause the affected nodes to go through the stop sequence which would result in the node sending a message to all other nodes in the cluster informing of them that it is leaving the cluster. This information can be used, and is used, by the membership algorithm to optimize certain aspects of the membership. (b)Involuntarily Leaves: There are two distinct cases for involuntarily leaves: i. The node can complete its abort or stop sequences and can
"clean-up" after itself. More importantly, as far as CMM is concerned, the node can send a message, the same one as the voluntarily leaves, that would inform other members of the cluster that this node will not be part of the cluster. The same optimizations that can be performed for the voluntarily leaving of a node can actually be implemented here. A node may leave the cluster due to a request from an application program with the appropriate privileges, ϋ. The node does not complete its abort sequence and panics the system. This is the most difficult of all failures to deal with and is usually detected by the absence of a heart-beat message from the failed node. This failure is un-distinguishable from a network failure in an asynchronous distributed system. 4.4 The Algorithm
The membership algorithm is described in this subsection. It relies on the assumptions and definitions described above. The user interface used in this algorithm is described later to make the flow of the algorithm "clean". Before we proceed to the description of the algorithm, we state the following rules that are enforced by our implementation of the membership algorithm;
• Each node, whether already part of the cluster or trying to join the cluster gets one and only one vote.
• Each node, /' will update its connectivity state matrix, C„ as soon as it hears from a node. The matrix C, is node 's understanding of the overall connectivity of the system. If node / does not hear from a nodey within the specified time, or is informed that node is down or unreachable, it will mark the e,, element of C, as zero. Furthermore, it will mark all the elements of theyth row as NULL, which implies that node / does not have any information about the connectivity of node /. For other rows of the matrix, node / will update them by replacing the kth row of its connectivity matrix with the connectivity vector V*- that it receives from node k.
• Each node /' will initially include its /th row of C, in its RECONF msg as the
MPr°P MPΓ°P proposed membership set, ' , that it is voting for. Note that the set '
Mprop proposed by node /' is different from the vector V,. ' is a proposed set that states a node's vote for other nodes in a binary form, whereas V, is a state vector
MP P which deals with the connectivity of the nodes in the system. Note that ' will have different elements, each element being a node id and a binary vote value, than V„ when nodes cannot agree on a stable membership and a subset of V, needs to be proposed as the new membership set.
• Each node keeps the total number of the nodes that are in its current view of the cluster membership, whether agreed or proposed, in a local variable N,. Note that N, is subject to the following rules during the execution of the membership algorithm:
Mprop (a)N, is initialized to the cardinality of . '
(b) N, is incremented for each node that is trying to join the cluster. (One increment per node as enforced via the nodeid check which is embedded in the message, done by the receiver thread.)
(c)N, is decremented for each node that aborts, as defined in part 2(b)i of Section 4.3, or voluntarily leaves. (Done by the receiver thread.)
(d) The quorum at the end of the membership algorithm is decided on this notion of N,.
• At the termination of the membership algorithm the nodes that form the cluster
M agreed agree on the new set of member nodes, ' . This set will be used in the next run through the membership algorithm, and at that time, all nodes that were
M agreed part of the previous configuration are assumed to have a consistent ' set.
• Each node, i, prior to entering the membership algorithm will have a sequence number, seq num, that is the same for all nodes in the current cluster. In addition each node will have its connectivity state matrix, C,. Note that C, is a n x n matrix, where n is the maximum number of nodes as defined by the current cluster configuration file, i.e. the current cdb file.
• Each node that is a joiner will have a variable, joining_node, set to TRUE. Once a
M agreed node has become a member of ' , it is no longer a joiner, and joining^node is set to FALSE.
• A node that is executing the first join will have a variable, start_cluster, set to TRUE. Nodes that are trying to join the cluster will have their start cluster variable initialized to FALSE.
All nodes of the cluster get their information about the various timeout values from the configuration file. Notations Tj, T2, ••• is used to denote possibly different timeout values. All these values need to be consistent among all nodes and are set to reasonable values which incorporate communication and queuing delays. The algorithm can be described by the following for each node, /':
membership_algorithm() { seq_num = seq_num + 1 ;
/* Broadcast a RECONF msg to all nodes proposing the set */
Mprop c in t
1* ' based on the initial connectivity matrix ' */
Mprop I* If a node is joining, its ' will contain only NULL elements. */
M agreed /* If the proposal does not include all of ' and /' is not a joiner, */
/* then all joiners are removed from the proposal. */
M Γ°P
1 = membershιp_proposal();
Retry:
While (!stable_proposal() && time < Ti) {
/* Node /' updates C„ N, and broadcast a RECONF msg *1
I* to all nodes in its V, as soon as /' gets and */ /* processes a new valid message. If message is from a node*/
/* that has not changed state and that / already has heard */
/* from, /' will not update C, and N, */
Mprop
1 = membership_proposal();
/* Nodes were unable to agree on a membership set. Therefore, i */
Mprop I* needs to modify its ' to reflect that an agreement may not be */
/* possible to reach. A new optimal subset will be proposed and */
/* broadcasted. */ propose_new membership();
While (!stable_proposal()) {
/* Node / will update C, and N, for any valid message and*/
/* will broadcast its RECONF msg. If a valid message */ /* indicates that some node has changed state, the variable */ /* node_state_changed is set to TRUE. */ if (node_state_changed) goto Retry;
}
/* If this is not the first join, we can check for quorum. if (!start_cluster) {
/* A joiner may not join a partially connected or nonexistent */
M agreed I* i cluster. */
Magreed Mprop if(joιning_node AND ( ' <F ' OR
M agreed V (/ e ' ), (ST, = DOWN)) abort_node();
/* Check to see if the proposed set has enough votes. Let v */
/* represent the number of votes that the current proposal has. */ if(v < L ( N,+ l)/2j) abort_node0; else if(v x 2 =N,) {
/* This is α possible split brain situation. */ if (N, = 2 && share_quorum_dev()) { if ( ! reserve_quorum()) abort_node();
} else wait_for_user_input(); } } /* A new membership has been agreed to. */
Magreed = Mprop
joining node = FALSE;
/* Node /' will broadcast another RECONF msg which will */
/* serve as a death message for all nodes that might have */
/* been stuck in a previous step of the membership */
/* algorithm. */
/* A variety of conditions must pass before t's proposal is considered
/* stable and an agreement has been reached. boolean stable_proposal() {
Mprop /* Not all sequence numbers of nodes in ' agree
Mprop if (V ( ' E ' ), seq_num, ≠ seq~num,!) return FALSE
/* Any state in C, is NULL if(V (s e s ≠NULL) return FALSE
Mf P is not still valid
Figure imgf000020_0001
return FALSE
/* All other nodes have not caught up to the local state yet
Mρrop i£(V0 e i ), STj ≠ STJ return FALSE
/* The proposals do not match
.e, , MP P , MPΦ MP P , iϊ(V(j e i ), J ≠ i ) return FALSE
return TRUE }
propose_new_membersbip() {
MP P vector proposal_to_test = ' ;
M agreed if (proposal_to_test ' AND fijlly_connected(proposa_to~test)) {
/* exclude joiners */ find_optimal_proposal(proposal_to_test)
else if (joining_node) { abort-nodeO; else
/* exclude joiners from proposal */ find_optimal_proposal(proposal_to_test & \3)>'
}
}
}
find_optimal_proposal(test_proposal) { while ( ! fully_connected(test_proposal)) test_proposal = get_next_proposal(test_proposal)
}
Figure imgf000021_0001
abort_node()
}
get_next_roposal(orignal) {
/* This routine follows rules 3 & 4 in Section 4.2 to select */
/* cluster proposals in order of decreasing optimality, based on the */ /* original set (original) from which we are selecting. */
/* We exhaustively cycle through all combinations of nodes in */
/* decreasing order of number of nodes. The combinations are */
/* ordered with lower node ids having precedence over higher */
/* node ids. */
}
Note that in the above algorithm we assume that there is a way to send messages to all nodes that are part of the cluster. If a node is down or unreachable, this is assumed to have been dealt with in previous reconfigurations and reflected in the matrix c in 'it . Note that the concept of a valid message was discussed above. Furthermore, the new optimal subset in the above algorithm is selected according to the rules and definitions described above. Also note that the entire algorithm is being executed within the begin step of the reconfiguration which is a timed event with its own timeout value.
In the above algorithm, the function membership_proposal() returns a membership proposal based on C/, including all nodes that are not in the DOWN state. It also excludes
M agreed all joiners from the proposal if the proposal does not include all of ' . An important function is the stable_proposal() function. This function decides if the proposed
Mprop set ' is agreed upon by all the members of that set. In order to count the number of
. MPΓ°P . . votes, node / needs to compare the proposed set from other nodes, i.e. J with its
Mprop own ' , j ≠ i . Note that the fijnction share_quorum_dev() is implemented by using the CCD dynamic file and informs the membership algorithm of the cases, such as a two node cluster, in which two nodes do share a quorum device. The binary function reserve_quorumQ returns false if and only if the device is already reserved by another node. The function ait_for_user_input() is discussed in greater detail below.
The function propose_new_membership() will get called to find an optimal subset
Mprop of ' , according to optimality conditions described above. It exhaustively tests the
Mprop subset combinations of ' until the first fully-connected set is found. The fully connected(prop) function returns true if the candidate proposal prop is contained in
MP P all the proposals of the members of prop. Note that if ' is already fully-connected,
MPΓ°P the proposal will not change. Also note that ' is not fully-connected, no joiners will be in the proposal. Finally, the fϊnd_optimal_proposal() and get_next_proposaI() functions implement the exhaustive search. 4.5 User Interfaces
Above, we deferred the discussion of how we will break the tie with the user input if there is a potential split-brain situation. In this subsection we specify how this is to be implemented.
The situations in which a set of nodes, X, and a different set of nodes, Fboth have exactly Nil votes, where N is the number of nodes in the previous cluster, is the case for which we will need operators assistant. Note that if the cardinality of both X and Fis one and they do share a quorum device, then we do not need to solicit input from the operator. In both situations the node will be waiting for the user input by executing the wait_for_user_input() call in the membership algorithm. The call to wait_for_user_input() will cause the creation of a print thread that would continuously print a message informing the operator that he needs to break the potential tie. The message would identify, for the appropriate nodes, the sets X or Y that must be shut down or informed to stay up. The operator will break the tie by issuing the command pdbadmin stopnode to one set of nodes while issuing the new command pdbadmin continue to the other set. The set that receives the stop command will abort, while the other set will stop printing the messages and continue its reconfiguration. Alternatively, the operator can issue a clustm reconfigure command, which is a valid option if there was a communication break down and the operator has fixed it. Issuing a clustm reconfigure command will cause a new reconfiguration to take place. If the operator issues any other command, besides pdbadmin stopnode, clustm reconfigure or pdbadmin continue at this time, the command reader thread, will not signal the transitions thread that is waiting for one of those commands and will simply ignore the command. The print thread, meanwhile, will be printing these messages continuously once every few seconds, to inform the operator that some immediate action is required.
The function wait_for_user_input(), which is executed by the transitions thread, is implemented as follows:
wait_for_user_inputO
{
/* create the print thread and make it print */ /* an informative message every few seconds. */ cond_wait(&state_change_cv, &autm_lock); check_stop_abort(); /* remove the print thread. */ }
The above sequence of actions will cause the transitions thread to sleep on the condition variable state_change_cv which is flagged under the following conditions:
User issues a continue command.
User issues a stopnode command.
User issues an abort command.
User forces reconfiguration.
Node receives a message that indicates that a remote node in its current membership set has gone down.
Node has not received a message from a remote node in its current membership set for node down timeout.
All of these actions are suitable for flagging the transitions thread and allow the user to issue the right set of commands to ensure that only one primary group remains in operation in the cluster.
5 Failure Fencing and Resource Migration
Another component of the system that requires modification due to the new architecture is the failure fencing mechanism that is employed in some distributed computer systems. In this section, we discuss a solution to the general problem of resource migration and the specific problem of failure fencing. The solution provided is generic in the sense that it handles the various array topologies-cascaded, n + 1, cross-connected, and others-as well as different software configurations-CVM with Netdisk, stand alone VxVM, or others. It also handles the 2-node cross connected array configuration without treating it as a special case.
The assumptions and the general solution are discussed next. This is followed by a short note on how this also solves the resource migration problem - the migration of highly available disk groups, HA/NFS file systems, logical IP addresses on the public networks, etc.
5.1 Assumptions
In the case of a shared disk configuration with CVM and Netdisk, it is assumed that the master and its backup node for all NetDisk devices have direct physical access to the underlying physical device.
In the case of a shared nothing configuration with VxVM, it is assumed that each node that has primary ownership of a set of disk groups has direct physical access to the devices belonging to those disk groups. In more concrete terms, if Node N has primary ownership of a set of disk-groups G, all disks belonging to the disk groups in G can be found in the set of storage devices denoted by D that are connected to N.
We also assume that information about the primary and backup ownership of a NetDisk device or other resources is maintained in the Cluster Configuration Database, CCD, and is available to all nodes in a consistent manner. We enforce this assumption by using the dynamic portion of the CCD. In particular, it is important that the CCD can be queried to obtain this information when the steps for failure fencing and resource migration gets executed during the reconfiguration process as outlined in the next subsection. These steps are executed only after cluster membership has been determined and quorum has been obtained.
5.2 Failure Fencing
In some distributed computer systems, every node has a backup node. A node (primary) and its backup node share a set of common devices they are connected to. This is denoted by B(Nj) = Nj. In the case of a CVM plus Netdisk configuration, the backup node becomes the master of the set of NetDisk devices owned by a failed node. In the case of VxVM configuration, the backup node becomes the primary owner of the set of disk group resources owned by a failed node.
Let N, denote a node of a cluster and D, denote a set of storage devices (composed of one or more SSA and/or Multipacks). Assuming that there are four nodes in the cluster, Nj to Nφ we will have the following relations for a cascaded configuration: B(Nj) = N4, B(N2) = Nj, BfN ) = N » BfN^) = N3. For the n + 1 configuration the relation is given by: B(N ) = N4, B(N2) = N4, B(Ns) = N4. Note that in this case node N does not have a backup node. And, finally for the cross connected case, the backup and primary relation is given by: B(N ) = N2, (N ) = Nj, B(Ns) = N4, B(N4) = N3. Note that the two-node cross connected case simply reduces to the case of: B(N ) = N2, B(N2) = Nj.
5.3 Generic Solution
Assume node / has failed and all the other nodes undergo a reconfiguration as a result of this failure. Each surviving node / will execute the following simple step after membership and quorum has been determined: if (B(Nj) = Nj) /* i.e. my nodeid */ { if (N masters NetDisk devices) /* for CVM & NetDisk */ { take over these devices and issue reservations on the corresponding physical devices;
} if (N, has primary ownership of disk groups) /* for VxVM */ { take ownership of these disk groups; issue reservations on the corresponding physical devices of the disk group:
} }
if (B(Nβ = Ni) N) /* i.e. the failed node is my backup */ {
/* I need to protect all my shared devices */ For all NetDisk devices for which N is the master issue reservations on corresponding physical devices; For all disk groups for which N, is the primary owner issue reservations on the corresponding physical devices; }
Note that "takeover these devices" implies that whatever interface NetDisk provides to takeover ownership of NetDisk devices will be used to accomplish this.
Without going into the syntactic details of how exactly a node determines what NetDisk devices are mastered by a failed node, it is sufficient to state that CCD maintains this information in its database and can be queried to obtain this information and. also whether the currently reconfiguring node is the backup of the failed node.
In some distributed computer systems, information about the primary ownership of a disk group is maintained in the cdb file in the following format:
cluster.node.().cdg: dgl dg2 cluster.nodc.l.cdg: dg3 dg4.
It should be a simple matter to find an equivalent representation for this information to be placed in the CCD and made available to all nodes in exactly the same manner as the NetDisk device configuration. The extra representation to be added for each node is of course the backup node in the same manner as NetDisk devices. For example:
cdg: dgl, dg2: 0.1.
The primary owner of cluster disk-groups dgl and dg2 is node 0 and its backup node is node 1.
It is also possible to either query CCD or the volume manager to find out the set of physical devices associated with a particular NetDisk virtual device or a particular disk group, respectively.
Finally, these are the steps executed when a failed node /' is ready to join the cluster. Each of the other nodes ≠ / execute this sequence in some as yet undetermined step k of the reconfiguration process:
IΪB(N) =Nj
Release reservations on NetDisk devices for which N/ is master
Use NetDisk interface to switch NetDisk devices mastered on Node N/ from N/ to Nj. Release reservations on disks in disk groups for which Nj is primary owner.
Deport the disk groups owned by Nj.
' Fi
lϊB(Nj) = Nj /* node that is joining is mv backup */
Release reservations on all NetDisk devices mastered on Nj.
Release reservations on all devices in disk groups whose primary owner is Nj.
Fi
The following sequence is executed by the joining Node Nj, in step k ÷ 1 of the reconfiguration process:
Master NetDisk devices mastered on Node Nj. Import disk groups for which N;- is primary owner.
For a node / ≠ /, it may not be possible to figure out from the membership whether i is joining the cluster or just undergoing a reconfiguration and was already part of the cluster. This does not matter as it is a simple matter to figure out whether node N, owns resources whose primary owners are part of the cluster membership and have it undertake the appropriate actions. If the algorithms are implemented correctly, at no point in time should node N, own resources belonging to node N, if N, was already part of the cluster. This is a safe assumption required for correctness and integrity. This algorithm is slightly expensive in terms of reconfiguration times, but in no way it constitutes a bottleneck.
We can utilize the approach of this section to solve the general resource migration problem in certain distributed computer systems. Resources that need to be highly available, are migrated over to a surviving node from a failed one. Examples of such resources are disk groups in a shared nothing database environment, disk groups for HA- NFS file-systems and, logical IP addresses. Arbitrary resources can be designated in the CCD with a master and a backup node. For example, the logical IP addresses can be migrated from failed nodes to surviving ones. Note that for a switch-over to take place, the backup node would have to release the resources of the joining node one step before the joining node takes over its resources.
5.4 A Restriction on Disk Groups
For some distributed computer systems, it is not possible to have arbitrary connections to the arrays from the nodes of the cluster. This is because, a disk group that is scattered across several arrays cannot be migrated to different nodes of the cluster, but needs to be migrated in its entirety to a single node. To illustrate, consider a configuration in which there are four nodes, Nj, ..., N4, and four array devices. D D4. Assume that node Nj has a physical connection to array Dj. In addition, N2 is physically connected to Dj and D3 and N3 is physically connected to D2 and D4. Finally, assume that Nj and N4 have no further connections.
Let's say the node N2 has a disk group G whose devices are scattered on disks in arrays D and D3. lfN2 now fails, G cannot be imported in its entirety on either Nj or N3 since all of its disks won't be visible on either N or N3. Such configurations are not supported in some distributed computer systems. If a node owns a disk group and if the node fails, it should be possible for the disk group in its entirety to be taken over by one of the surviving nodes. This does not constrain the array topology, but places restraints on how data is scattered across the arrays.
5.5 A Migration Strategy with Minimal Effort
One of the more time consuming activities in a system is laying out of the data. We propose to miniinize this for the those who wish to upgrade their existing two node clusters to three nodes, or those with a three node cluster that wish to update their clusters to four nodes. We are not going to do this dynamically. The cluster will be shut down and restarted. The only criteria is to allow access to the mirror and the primary copy of the data from the same node without having to relay all the volumes and/or disk groups. This may require addition of adaptor cards.. The above description is illustrative only and is not limiting. The present invention is therefore defined solely and completely by the appended claims together with their full scope of equivalents.

Claims

What is claimed is:
1. A method for determining membership of nodes in a distributed computer system, the method comprising:
(a) determining connection data representing interconnectivity of the nodes of the distributed computer system;
(b) applying optimization criteria to the connection data to form a proposed membership list of a proposed new cluster;
(c) broadcasting the proposed membership list to the connected nodes;
(d) receiving other proposed membership lists from the connected nodes; and
(e) selecting an elected proposed membership list from the other proposed membership lists.
2. The method of Claim 1 wherein (a) determining connection data comprises: determining with which connected ones of other nodes of the distributed computer system a selected node is in communication; broadcasting to the other nodes data specifying the connected nodes; receiving node connection data from the connected nodes; and combining the node connection data from the connected nodes and the data specifying the connected nodes to form the connection data.
3. The method of Claim 1 wherein (e) selecting an elected proposed membership list comprises: ensuring that the proposed membership list and all other proposed membership lists agree.
4. The method of Claim 3 wherein (e) selecting an elected proposed membership list further comprises: detecting disagreement between the proposed membership list and all other proposed membership lists; and in response to such disagreement, repeating (a) through (d).
' 5. The method of Claim 1 wherein (e) selecting an elected proposed membership list comprises: determining that the elected proposed membership list represents nodes which collectively form a quorum.
6. The method of Claim 5 wherein determining that the elected proposed membership list represents nodes which collectively form a quorum comprises: estimating a number of nodes of the distributed computer system are operational.
7. The method of Claim 6 wherein estimating a number of nodes of the distributed computer system are operational comprises: determining a number of nodes represented in the first-mentioned proposed membership list; adding a number of joining nodes; and subtracting a number of voluntarily leaving nodes.
8. A computer readable medium useful in association with a computer which includes a processor and a memory, the computer readable medium including computer instructions which are configured to cause the computer to determine membership of nodes in a distributed computer system by:
(a) deteπnining connection data representing interconnectivity of the nodes of the distributed computer system;
(b) applying optimization criteria to the connection data to form a proposed membership list of a proposed new cluster;
(c) broadcasting the proposed membership list to the connected nodes;
(d) receiving other proposed membership lists from the connected nodes; and
(e) selecting an elected proposed membership list from the other proposed membership lists.
9. The computer readable medium of Claim 8 wherein (a) determining connection data comprises: determining with which connected ones of other nodes of the distributed computer system a selected node is in communication; broadcasting to the other nodes data specifying the connected nodes; receiving node connection data from the connected nodes; and combining the node connection data from the connected nodes and the data specifying the connected nodes to form the connection data.
10. The computer readable medium of Claim 8 wherein (e) selecting an elected proposed membership list comprises: ensuring that the proposed membership list and all other proposed membership lists agree.
11. The computer readable medium of Claim 10 wherein (e) selecting an elected proposed membership list further comprises: detecting disagreement between the proposed membership list and all other proposed membership lists; and in response to such disagreement, repeating (a) through (d).
12. The computer readable medium of Claim 8 wherein (e) selecting an elected proposed membership list comprises: determining that the elected proposed membership list represents nodes which collectively form a quorum.
13. The computer readable medium of Claim 12 wherein determining that the elected proposed membership list represents nodes which collectively form a quorum comprises: estimating a number of nodes of the distributed computer system are operational.
14. The computer readable medium of Claim 13 wherein estimating a number of nodes of the distributed computer system are operational comprises: determining a number of nodes represented in the first-mentioned proposed ' membership list; adding a number of joining nodes; and subtracting a number of voluntarily leaving nodes.
15. A computer system comprising: a processor; a memory operatively coupled to the processor; and a failure detection module (i) which executes in the processor from the memory and (ϋ) which, when executed by the processor, causes the computer to determine membership of nodes in a distributed computer system by:
(a) determining connection data representing interconnectivity of the nodes of the distributed computer system;
(b) applying optimization criteria to the connection data to form a proposed membership list of a proposed new cluster;
(c) broadcasting the proposed membership list to the connected nodes;
(d) receiving other proposed membership lists from the connected nodes; and
(e) selecting an elected proposed membership list from the other proposed membership lists.
16. The computer system of Claim 15 wherein (a) determining connection data comprises: determining with which connected ones of other nodes of the distributed computer system a selected node is in communication; broadcasting to the other nodes data specifying the connected nodes; receiving node connection data from the connected nodes; and combining the node connection data from the connected nodes and the data specifying the connected nodes to form the connection data.
17. The computer system of Claim 15 wherein (e) selecting an elected proposed membership list comprises: ensuring that the proposed membership list and all other proposed ' membership lists agree.
18. The computer system of Claim 17 wherein (e) selecting an elected proposed membership list further comprises: detecting disagreement between the proposed membership list and all other proposed membership lists; and in response to such disagreement, repeating (a) through (d).
19. The computer system of Claim 15 wherein (e) selecting an elected proposed membership list comprises: determining that the elected proposed membership list represents nodes which collectively form a quorum.
20. The computer system of Claim 19 wherein determining that the elected proposed membership list represents nodes which collectively form a quorum comprises: estimating a number of nodes of the distributed computer system are operational.
21. The computer system of Claim 20 wherein estimating a number of nodes of the distributed computer system are operational comprises: determining a number of nodes represented in the first-mentioned proposed membership list; adding a number of joining nodes; and subtracting a number of voluntarily leaving nodes.
PCT/US1998/022161 1997-10-21 1998-10-20 Determining cluster membership in a distributed computer system WO1999021098A2 (en)

Priority Applications (4)

Application Number Priority Date Filing Date Title
CA002306718A CA2306718A1 (en) 1997-10-21 1998-10-20 Determining cluster membership in a distributed computer system
EP98953773A EP1025506A2 (en) 1997-10-21 1998-10-20 Determining cluster membership in a distributed computer system
JP2000517348A JP2001521222A (en) 1997-10-21 1998-10-20 Method for determining cluster membership in a distributed computer system
AU11054/99A AU1105499A (en) 1997-10-21 1998-10-20 Determining cluster membership in a distributed computer system

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
US08/955,885 US5999712A (en) 1997-10-21 1997-10-21 Determining cluster membership in a distributed computer system
US08/955,885 1997-10-21

Publications (2)

Publication Number Publication Date
WO1999021098A2 true WO1999021098A2 (en) 1999-04-29
WO1999021098A3 WO1999021098A3 (en) 1999-07-01

Family

ID=25497484

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/US1998/022161 WO1999021098A2 (en) 1997-10-21 1998-10-20 Determining cluster membership in a distributed computer system

Country Status (6)

Country Link
US (2) US5999712A (en)
EP (1) EP1025506A2 (en)
JP (1) JP2001521222A (en)
AU (1) AU1105499A (en)
CA (1) CA2306718A1 (en)
WO (1) WO1999021098A2 (en)

Cited By (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2000058825A2 (en) * 1999-03-26 2000-10-05 Microsoft Corporation Data distribution in a server cluster
EP1117041A2 (en) * 2000-01-10 2001-07-18 Sun Microsystems, Inc. Method and apparatus for managing failures in clustered computer systems
EP1117039A2 (en) 2000-01-10 2001-07-18 Sun Microsystems, Inc. Controlled take over of services by remaining nodes of clustered computing system
EP1134658A2 (en) * 2000-03-14 2001-09-19 Sun Microsystems, Inc. A system and method for comprehensive availability management in a high-availability computer system
EP1877901A2 (en) * 2005-05-06 2008-01-16 Marathon Technologies Corporation Fault tolerant computer system
JP2015535970A (en) * 2012-09-07 2015-12-17 アビジロン コーポレイション Physical security system having multiple server nodes
US9959109B2 (en) 2015-04-10 2018-05-01 Avigilon Corporation Upgrading a physical security system having multiple server nodes
US20230229361A1 (en) * 2020-08-14 2023-07-20 Inspur Suzhou Intelligent Technology Co., Ltd. Cluster arbitration method and system based on heterogeneous storage, and device and storage medium

Families Citing this family (195)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5999712A (en) * 1997-10-21 1999-12-07 Sun Microsystems, Inc. Determining cluster membership in a distributed computer system
US6279032B1 (en) * 1997-11-03 2001-08-21 Microsoft Corporation Method and system for quorum resource arbitration in a server cluster
US6092220A (en) * 1997-11-17 2000-07-18 International Business Machines Corporation Method and apparatus for ordered reliable multicast with asymmetric safety in a multiprocessing system
US6421787B1 (en) * 1998-05-12 2002-07-16 Sun Microsystems, Inc. Highly available cluster message passing facility
US6243744B1 (en) * 1998-05-26 2001-06-05 Compaq Computer Corporation Computer network cluster generation indicator
US6311217B1 (en) * 1998-06-04 2001-10-30 Compaq Computer Corporation Method and apparatus for improved cluster administration
US6438582B1 (en) * 1998-07-21 2002-08-20 International Business Machines Corporation Method and system for efficiently coordinating commit processing in a parallel or distributed database system
US6192443B1 (en) * 1998-07-29 2001-02-20 International Business Machines Corporation Apparatus for fencing a member of a group of processes in a distributed processing environment
US6212595B1 (en) * 1998-07-29 2001-04-03 International Business Machines Corporation Computer program product for fencing a member of a group of processes in a distributed processing environment
US6205510B1 (en) * 1998-07-29 2001-03-20 International Business Machines Corporation Method for fencing a member of a group of processes in a distributed processing environment
US6687754B1 (en) * 1998-08-27 2004-02-03 Intel Corporation Method of detecting a device in a network
US6275847B1 (en) * 1999-01-07 2001-08-14 Iq Net Solutions, Inc. Distributed processing systems incorporating processing zones which communicate according to both streaming and event-reaction protocols
US6424990B1 (en) 1999-01-07 2002-07-23 Jeffrey I. Robinson Distributed processing systems incorporating a plurality of cells which process information in response to single events
US6272524B1 (en) * 1999-01-07 2001-08-07 Iq Netsolutions Inc. Distributed processing systems incorporating a plurality of cells which process information in response to single events
US6272526B1 (en) * 1999-01-07 2001-08-07 Iq Netsolutions, Inc. Distributed processing systems having self-advertising cells
US6272527B1 (en) * 1999-01-07 2001-08-07 Iq Net Solutions, Inc. Distributed processing systems incorporating nodes having processing cells which execute scripts causing a message to be sent internodally
US6272525B1 (en) * 1999-01-07 2001-08-07 Iq Net Solutions, Inc. Distributed processing systems including processing zones which subscribe and unsubscribe to mailing lists
US6401120B1 (en) 1999-03-26 2002-06-04 Microsoft Corporation Method and system for consistent cluster operational data in a server cluster using a quorum of replicas
US7774469B2 (en) * 1999-03-26 2010-08-10 Massa Michael T Consistent cluster operational data in a server cluster using a quorum of replicas
US6745241B1 (en) * 1999-03-31 2004-06-01 International Business Machines Corporation Method and system for dynamic addition and removal of multiple network names on a single server
US6968390B1 (en) 1999-04-15 2005-11-22 International Business Machines Corporation Method and system for enabling a network function in a context of one or all server names in a multiple server name environment
US6502203B2 (en) * 1999-04-16 2002-12-31 Compaq Information Technologies Group, L.P. Method and apparatus for cluster system operation
US7020695B1 (en) 1999-05-28 2006-03-28 Oracle International Corporation Using a cluster-wide shared repository to provide the latest consistent definition of the cluster (avoiding the partition-in time problem)
US6532494B1 (en) * 1999-05-28 2003-03-11 Oracle International Corporation Closed-loop node membership monitor for network clusters
US6871222B1 (en) 1999-05-28 2005-03-22 Oracle International Corporation Quorumless cluster using disk-based messaging
US7076783B1 (en) * 1999-05-28 2006-07-11 Oracle International Corporation Providing figure of merit vote from application executing on a partitioned cluster
US6490693B1 (en) * 1999-08-31 2002-12-03 International Business Machines Corporation Dynamic reconfiguration of a quorum group of processors in a distributed computing system
US7464147B1 (en) * 1999-11-10 2008-12-09 International Business Machines Corporation Managing a cluster of networked resources and resource groups using rule - base constraints in a scalable clustering environment
US6745240B1 (en) * 1999-11-15 2004-06-01 Ncr Corporation Method and apparatus for configuring massively parallel systems
US6662219B1 (en) * 1999-12-15 2003-12-09 Microsoft Corporation System for determining at subgroup of nodes relative weight to represent cluster by obtaining exclusive possession of quorum resource
US6658470B1 (en) * 1999-12-17 2003-12-02 International Business Machines Corporation Centralized logging of global reliability, availability, and serviceability (GRAS) services data for a distributed environment and backup logging system and method in event of failure
US6769008B1 (en) * 2000-01-10 2004-07-27 Sun Microsystems, Inc. Method and apparatus for dynamically altering configurations of clustered computer systems
US6920454B1 (en) * 2000-01-28 2005-07-19 Oracle International Corporation Techniques for DLM optimization with transferring lock information
US7246120B2 (en) 2000-01-28 2007-07-17 Oracle International Corporation Techniques for achieving higher availability of resources during reconfiguration of a cluster
US6751616B1 (en) 2000-01-28 2004-06-15 Oracle International Corp. Techniques for DLM optimization with re-mapping responsibility for lock management
US6636982B1 (en) * 2000-03-03 2003-10-21 International Business Machines Corporation Apparatus and method for detecting the reset of a node in a cluster computer system
US20020198996A1 (en) 2000-03-16 2002-12-26 Padmanabhan Sreenivasan Flexible failover policies in high availability computing systems
US6775703B1 (en) * 2000-05-01 2004-08-10 International Business Machines Corporation Lease based safety protocol for distributed system with multiple networks
US6847993B1 (en) 2000-05-31 2005-01-25 International Business Machines Corporation Method, system and program products for managing cluster configurations
US6725261B1 (en) * 2000-05-31 2004-04-20 International Business Machines Corporation Method, system and program products for automatically configuring clusters of a computing environment
US6801937B1 (en) * 2000-05-31 2004-10-05 International Business Machines Corporation Method, system and program products for defining nodes to a cluster
US7325046B1 (en) * 2000-05-31 2008-01-29 International Business Machines Corporation Method, system and program products for managing processing groups of a distributed computing environment
US6807557B1 (en) * 2000-05-31 2004-10-19 International Business Machines Corporation Method, system and program products for providing clusters of a computing environment
US6968359B1 (en) * 2000-08-14 2005-11-22 International Business Machines Corporation Merge protocol for clustered computer system
US7113995B1 (en) * 2000-10-19 2006-09-26 International Business Machines Corporation Method and apparatus for reporting unauthorized attempts to access nodes in a network computing system
US6886038B1 (en) * 2000-10-24 2005-04-26 Microsoft Corporation System and method for restricting data transfers and managing software components of distributed computers
US7606898B1 (en) * 2000-10-24 2009-10-20 Microsoft Corporation System and method for distributed management of shared computers
US7093288B1 (en) 2000-10-24 2006-08-15 Microsoft Corporation Using packet filters and network virtualization to restrict network communications
US6915338B1 (en) * 2000-10-24 2005-07-05 Microsoft Corporation System and method providing automatic policy enforcement in a multi-computer service application
US6907395B1 (en) 2000-10-24 2005-06-14 Microsoft Corporation System and method for designing a logical model of a distributed computer system and deploying physical resources according to the logical model
US7113900B1 (en) 2000-10-24 2006-09-26 Microsoft Corporation System and method for logical modeling of distributed computer systems
US6839752B1 (en) 2000-10-27 2005-01-04 International Business Machines Corporation Group data sharing during membership change in clustered computer system
US7185099B1 (en) 2000-11-22 2007-02-27 International Business Machines Corporation Apparatus and method for communicating between computer systems using a sliding send window for ordered messages in a clustered computing environment
US7769844B2 (en) 2000-12-07 2010-08-03 International Business Machines Corporation Peer protocol status query in clustered computer system
US7035938B2 (en) * 2000-12-07 2006-04-25 Telcordia Technologies, Inc. Determination of connection links to configure a virtual private network
US7502857B2 (en) * 2000-12-15 2009-03-10 International Business Machines Corporation Method and system for optimally allocating a network service
US6785678B2 (en) * 2000-12-21 2004-08-31 Emc Corporation Method of improving the availability of a computer clustering system through the use of a network medium link state function
US7792977B1 (en) * 2001-02-28 2010-09-07 Oracle International Corporation Method for fencing shared resources from cluster nodes
US6952766B2 (en) * 2001-03-15 2005-10-04 International Business Machines Corporation Automated node restart in clustered computer system
US7305450B2 (en) * 2001-03-29 2007-12-04 Nokia Corporation Method and apparatus for clustered SSL accelerator
US6918051B2 (en) * 2001-04-06 2005-07-12 International Business Machines Corporation Node shutdown in clustered computer system
US6675264B2 (en) * 2001-05-07 2004-01-06 International Business Machines Corporation Method and apparatus for improving write performance in a cluster-based file system
US7617292B2 (en) 2001-06-05 2009-11-10 Silicon Graphics International Multi-class heterogeneous clients in a clustered filesystem
US7640582B2 (en) 2003-04-16 2009-12-29 Silicon Graphics International Clustered filesystem for mix of trusted and untrusted nodes
US20040139125A1 (en) * 2001-06-05 2004-07-15 Roger Strassburg Snapshot copy of data volume during data access
US8010558B2 (en) 2001-06-05 2011-08-30 Silicon Graphics International Relocation of metadata server with outstanding DMAPI requests
US7016946B2 (en) * 2001-07-05 2006-03-21 Sun Microsystems, Inc. Method and system for establishing a quorum for a geographically distributed cluster of computers
US6880100B2 (en) * 2001-07-18 2005-04-12 Smartmatic Corp. Peer-to-peer fault detection
US20030028594A1 (en) * 2001-07-31 2003-02-06 International Business Machines Corporation Managing intended group membership using domains
US6925582B2 (en) * 2001-08-01 2005-08-02 International Business Machines Corporation Forwarding of diagnostic messages in a group
US7239606B2 (en) * 2001-08-08 2007-07-03 Compunetix, Inc. Scalable configurable network of sparsely interconnected hyper-rings
US7243374B2 (en) 2001-08-08 2007-07-10 Microsoft Corporation Rapid application security threat analysis
DE10143142A1 (en) * 2001-09-04 2003-01-30 Bosch Gmbh Robert Microprocessor-controlled operation of vehicular EEPROM memory, employs two memory areas with data pointers and cyclic validation strategy
US7231461B2 (en) * 2001-09-14 2007-06-12 International Business Machines Corporation Synchronization of group state data when rejoining a member to a primary-backup group in a clustered computer system
US7277952B2 (en) * 2001-09-28 2007-10-02 Microsoft Corporation Distributed system resource protection via arbitration and ownership
US7350098B2 (en) * 2001-11-30 2008-03-25 Oracle International Corporation Detecting events of interest for managing components on a high availability framework
US20030177213A1 (en) * 2002-01-18 2003-09-18 Wallace Chris E. Determining connectivity in communication networks
US6950855B2 (en) * 2002-01-18 2005-09-27 International Business Machines Corporation Master node selection in clustered node configurations
US7240088B2 (en) * 2002-01-25 2007-07-03 International Business Machines Corporation Node self-start in a decentralized cluster
US7203748B2 (en) * 2002-02-15 2007-04-10 International Business Machines Corporation Method for detecting the quick restart of liveness daemons in a distributed multinode data processing system
US8321543B2 (en) * 2002-03-04 2012-11-27 International Business Machines Corporation System and method for determining weak membership in set of computer nodes
US7631066B1 (en) * 2002-03-25 2009-12-08 Symantec Operating Corporation System and method for preventing data corruption in computer system clusters
US7379970B1 (en) * 2002-04-05 2008-05-27 Ciphermax, Inc. Method and system for reduced distributed event handling in a network environment
US7051102B2 (en) * 2002-04-29 2006-05-23 Microsoft Corporation Peer-to-peer name resolution protocol (PNRP) security infrastructure and method
US7093010B2 (en) * 2002-05-20 2006-08-15 Telefonaktiebolaget Lm Ericsson (Publ) Operator-defined consistency checking in a network management system
US6925541B2 (en) * 2002-06-12 2005-08-02 Hitachi, Ltd. Method and apparatus for managing replication volumes
US7590985B1 (en) 2002-07-12 2009-09-15 3Par, Inc. Cluster inter-process communication transport
US6965957B1 (en) * 2002-07-12 2005-11-15 3Pardata, Inc. Automatic cluster join protocol
US20040027155A1 (en) * 2002-08-08 2004-02-12 Schlansker Michael S. System and method for self configuration of reconfigurable systems
JP3800158B2 (en) * 2002-09-27 2006-07-26 ブラザー工業株式会社 Data transmission system, terminal device, and program
US20040181707A1 (en) * 2003-03-11 2004-09-16 Hitachi, Ltd. Method and apparatus for seamless management for disaster recovery
US7447786B2 (en) 2003-05-09 2008-11-04 Oracle International Corporation Efficient locking of shared data that is accessed for reads in a cluster database
US7085897B2 (en) * 2003-05-12 2006-08-01 International Business Machines Corporation Memory management for a symmetric multiprocessor computer system
US7376724B2 (en) * 2003-05-30 2008-05-20 Oracle International Corporation Dynamic reconfiguration of nodes in a cluster file system
US7467168B2 (en) * 2003-06-18 2008-12-16 International Business Machines Corporation Method for mirroring data at storage locations
US7562154B2 (en) * 2003-06-30 2009-07-14 International Business Machines Corporation System and method for filtering stale messages resulting from membership changes in a distributed computing environment
US7739541B1 (en) * 2003-07-25 2010-06-15 Symantec Operating Corporation System and method for resolving cluster partitions in out-of-band storage virtualization environments
US7370101B1 (en) 2003-12-19 2008-05-06 Sun Microsystems, Inc. Automated testing of cluster data services
US7165189B1 (en) * 2003-12-19 2007-01-16 Sun Microsystems, Inc. Distributed test framework for clustered systems
US8005888B2 (en) * 2003-12-30 2011-08-23 Microsoft Corporation Conflict fast consensus
US7379952B2 (en) * 2004-01-30 2008-05-27 Oracle International Corporation Techniques for multiple window resource remastering among nodes of a cluster
JP2005310243A (en) * 2004-04-20 2005-11-04 Seiko Epson Corp Memory controller, semiconductor integrated circuit apparatus, semiconductor apparatus, microcomputer, and electronic equipment
US20050268151A1 (en) * 2004-04-28 2005-12-01 Nokia, Inc. System and method for maximizing connectivity during network failures in a cluster system
US7856502B2 (en) * 2004-06-18 2010-12-21 Microsoft Corporation Cheap paxos
US7334154B2 (en) * 2004-06-18 2008-02-19 Microsoft Corporation Efficient changing of replica sets in distributed fault-tolerant computing system
US20050289228A1 (en) * 2004-06-25 2005-12-29 Nokia Inc. System and method for managing a change to a cluster configuration
US7590737B1 (en) 2004-07-16 2009-09-15 Symantec Operating Corporation System and method for customized I/O fencing for preventing data corruption in computer system clusters
EP1805947A1 (en) * 2004-09-29 2007-07-11 Telefonaktiebolaget LM Ericsson (publ) Installing a new view of a cluster membership
US20080201403A1 (en) * 2004-09-29 2008-08-21 Telefonaktiebolaget Lm Ericsson (Publ) Maintaning a View of a Cluster's Membership
US8185776B1 (en) * 2004-09-30 2012-05-22 Symantec Operating Corporation System and method for monitoring an application or service group within a cluster as a resource of another cluster
US20060074940A1 (en) * 2004-10-05 2006-04-06 International Business Machines Corporation Dynamic management of node clusters to enable data sharing
US8090880B2 (en) 2006-11-09 2012-01-03 Microsoft Corporation Data consistency within a federation infrastructure
US7694167B2 (en) * 2004-10-22 2010-04-06 Microsoft Corporation Maintaining routing consistency within a rendezvous federation
US7730220B2 (en) * 2004-10-22 2010-06-01 Microsoft Corporation Broadcasting communication within a rendezvous federation
US8392515B2 (en) * 2004-10-22 2013-03-05 Microsoft Corporation Subfederation creation and maintenance in a federation infrastructure
US20110082928A1 (en) * 2004-10-22 2011-04-07 Microsoft Corporation Maintaining consistency within a federation infrastructure
US20060090003A1 (en) * 2004-10-22 2006-04-27 Microsoft Corporation Rendezvousing resource requests with corresponding resources
US8095601B2 (en) 2004-10-22 2012-01-10 Microsoft Corporation Inter-proximity communication within a rendezvous federation
US8549180B2 (en) 2004-10-22 2013-10-01 Microsoft Corporation Optimizing access to federation infrastructure-based resources
US7958262B2 (en) * 2004-10-22 2011-06-07 Microsoft Corporation Allocating and reclaiming resources within a rendezvous federation
US8095600B2 (en) * 2004-10-22 2012-01-10 Microsoft Corporation Inter-proximity communication within a rendezvous federation
US8014321B2 (en) * 2004-10-22 2011-09-06 Microsoft Corporation Rendezvousing resource requests with corresponding resources
US20060200469A1 (en) * 2005-03-02 2006-09-07 Lakshminarayanan Chidambaran Global session identifiers in a multi-node system
US7209990B2 (en) * 2005-04-05 2007-04-24 Oracle International Corporation Maintain fairness of resource allocation in a multi-node environment
US20060242453A1 (en) * 2005-04-25 2006-10-26 Dell Products L.P. System and method for managing hung cluster nodes
US20070022314A1 (en) * 2005-07-22 2007-01-25 Pranoop Erasani Architecture and method for configuring a simplified cluster over a network with fencing and quorum
US8812501B2 (en) * 2005-08-08 2014-08-19 Hewlett-Packard Development Company, L.P. Method or apparatus for selecting a cluster in a group of nodes
US7941537B2 (en) * 2005-10-03 2011-05-10 Genband Us Llc System, method, and computer-readable medium for resource migration in a distributed telecommunication system
US7941309B2 (en) 2005-11-02 2011-05-10 Microsoft Corporation Modeling IT operations/policies
US7953890B1 (en) * 2006-01-27 2011-05-31 Symantec Operating Corporation System and method for switching to a new coordinator resource
US7979460B2 (en) 2006-02-15 2011-07-12 Sony Computer Entainment America Inc. Systems and methods for server management
GB0622553D0 (en) * 2006-11-11 2006-12-20 Ibm A method, apparatus or software for managing partitioning in a cluster of nodes
US7613947B1 (en) * 2006-11-30 2009-11-03 Netapp, Inc. System and method for storage takeover
JP4505763B2 (en) * 2007-01-31 2010-07-21 ヒューレット−パッカード デベロップメント カンパニー エル．ピー． Managing node clusters
US8209417B2 (en) * 2007-03-08 2012-06-26 Oracle International Corporation Dynamic resource profiles for clusterware-managed resources
US7890555B2 (en) * 2007-07-10 2011-02-15 International Business Machines Corporation File system mounting in a clustered file system
US8275866B2 (en) * 2007-11-13 2012-09-25 At&T Intellectual Property I, L.P. Assigning telecommunications nodes to community of interest clusters
US8706858B2 (en) * 2008-04-17 2014-04-22 Alcatel Lucent Method and apparatus for controlling flow of management tasks to management system databases
US20090265450A1 (en) * 2008-04-17 2009-10-22 Darren Helmer Method and apparatus for managing computing resources of management systems
US8892689B1 (en) * 2008-04-30 2014-11-18 Netapp, Inc. Method and apparatus for a storage server to automatically discover and join a network storage cluster
US8498647B2 (en) * 2008-08-28 2013-07-30 Qualcomm Incorporated Distributed downlink coordinated multi-point (CoMP) framework
WO2010044096A2 (en) * 2008-09-12 2010-04-22 Computational Research Laboratories Limited Cluster computing
US8443062B2 (en) * 2008-10-23 2013-05-14 Microsoft Corporation Quorum based transactionally consistent membership management in distributed storage systems
US8675511B2 (en) * 2008-12-10 2014-03-18 Qualcomm Incorporated List elimination for distributed downlink coordinated multi-point (CoMP) framework
KR101042908B1 (en) * 2009-02-12 2011-06-21 엔에이치엔(주) Method, system, and computer-readable recording medium for determining major group under split-brain network problem
US8903917B2 (en) * 2009-06-03 2014-12-02 Novell, Inc. System and method for implementing a cluster token registry for business continuity
CN101925183A (en) * 2009-06-15 2010-12-22 中兴通讯股份有限公司 Coordinated multi-point transmission based data transmission method and device
US8437282B2 (en) * 2009-06-21 2013-05-07 Clearone Communications Hong Kong Limited System and method of multi-endpoint data conferencing
US8108712B1 (en) * 2009-10-30 2012-01-31 Hewlett-Packard Development Company, L.P. Method and apparatus for removing a computer from a computer cluster observing failure
US8910176B2 (en) * 2010-01-15 2014-12-09 International Business Machines Corporation System for distributed task dispatch in multi-application environment based on consensus for load balancing using task partitioning and dynamic grouping of server instance
US8560365B2 (en) 2010-06-08 2013-10-15 International Business Machines Corporation Probabilistic optimization of resource discovery, reservation and assignment
US9646271B2 (en) 2010-08-06 2017-05-09 International Business Machines Corporation Generating candidate inclusion/exclusion cohorts for a multiply constrained group
US8370350B2 (en) 2010-09-03 2013-02-05 International Business Machines Corporation User accessibility to resources enabled through adaptive technology
US8968197B2 (en) 2010-09-03 2015-03-03 International Business Machines Corporation Directing a user to a medical resource
US8726274B2 (en) * 2010-09-10 2014-05-13 International Business Machines Corporation Registration and initialization of cluster-aware virtual input/output server nodes
US9292577B2 (en) 2010-09-17 2016-03-22 International Business Machines Corporation User accessibility to data analytics
US8429182B2 (en) 2010-10-13 2013-04-23 International Business Machines Corporation Populating a task directed community in a complex heterogeneous environment based on non-linear attributes of a paradigmatic cohort member
US9443211B2 (en) 2010-10-13 2016-09-13 International Business Machines Corporation Describing a paradigmatic member of a task directed community in a complex heterogeneous environment based on non-linear attributes
US8949828B2 (en) 2011-01-11 2015-02-03 International Business Machines Corporation Single point, scalable data synchronization for management of a virtual input/output server cluster
US9081839B2 (en) 2011-01-28 2015-07-14 Oracle International Corporation Push replication for use with a distributed data grid
US9164806B2 (en) 2011-01-28 2015-10-20 Oracle International Corporation Processing pattern framework for dispatching and executing tasks in a distributed computing grid
US9262229B2 (en) * 2011-01-28 2016-02-16 Oracle International Corporation System and method for supporting service level quorum in a data grid cluster
US9201685B2 (en) 2011-01-28 2015-12-01 Oracle International Corporation Transactional cache versioning and storage in a distributed data grid
US9063852B2 (en) 2011-01-28 2015-06-23 Oracle International Corporation System and method for use with a data grid cluster to support death detection
WO2013054425A1 (en) * 2011-10-13 2013-04-18 富士通株式会社 Node device and communication method
GB2496840A (en) 2011-11-15 2013-05-29 Ibm Controlling access to a shared storage system
US10706021B2 (en) 2012-01-17 2020-07-07 Oracle International Corporation System and method for supporting persistence partition discovery in a distributed data grid
US20140075170A1 (en) * 2012-09-12 2014-03-13 International Business Machines Corporation Automated firmware voting to enable multi-enclosure federated systems
US9619668B2 (en) * 2013-09-16 2017-04-11 Axis Ab Managing application data in distributed control systems
US9667496B2 (en) * 2013-12-24 2017-05-30 International Business Machines Corporation Configuration updates across peer storage systems
CN103995750B (en) * 2014-06-04 2017-03-22 重庆大学 Asymmetric distributed constrained optimization method for multi-Agent system
US9742692B2 (en) * 2014-06-23 2017-08-22 Microsoft Technology Licensing, Llc Acquiring resource lease using multiple lease servers
US10664495B2 (en) 2014-09-25 2020-05-26 Oracle International Corporation System and method for supporting data grid snapshot and federation
WO2016106682A1 (en) * 2014-12-31 2016-07-07 华为技术有限公司 Post-cluster brain split quorum processing method and quorum storage device and system
US10585599B2 (en) 2015-07-01 2020-03-10 Oracle International Corporation System and method for distributed persistent store archival and retrieval in a distributed computing environment
US10798146B2 (en) 2015-07-01 2020-10-06 Oracle International Corporation System and method for universal timeout in a distributed computing environment
US10860378B2 (en) 2015-07-01 2020-12-08 Oracle International Corporation System and method for association aware executor service in a distributed computing environment
US11163498B2 (en) 2015-07-01 2021-11-02 Oracle International Corporation System and method for rare copy-on-write in a distributed computing environment
US10341252B2 (en) * 2015-09-30 2019-07-02 Veritas Technologies Llc Partition arbitration optimization
WO2017149354A1 (en) * 2016-03-01 2017-09-08 Telefonaktiebolaget Lm Ericsson (Publ) Neighbor monitoring in a hyperscaled environment
US10049011B2 (en) * 2016-05-03 2018-08-14 International Business Machines Corporation Continuing operation of a quorum based system after failures
US10521344B1 (en) * 2017-03-10 2019-12-31 Pure Storage, Inc. Servicing input/output (‘I/O’) operations directed to a dataset that is synchronized across a plurality of storage systems
US11550820B2 (en) 2017-04-28 2023-01-10 Oracle International Corporation System and method for partition-scoped snapshot creation in a distributed data computing environment
US10459810B2 (en) 2017-07-06 2019-10-29 Oracle International Corporation Technique for higher availability in a multi-node system using replicated lock information to determine a set of data blocks for recovery
US10769019B2 (en) 2017-07-19 2020-09-08 Oracle International Corporation System and method for data recovery in a distributed data computing environment implementing active persistence
US10721095B2 (en) 2017-09-26 2020-07-21 Oracle International Corporation Virtual interface system and method for multi-tenant cloud networking
US10862965B2 (en) 2017-10-01 2020-12-08 Oracle International Corporation System and method for topics implementation in a distributed data computing environment
US10671494B1 (en) * 2017-11-01 2020-06-02 Pure Storage, Inc. Consistent selection of replicated datasets during storage system recovery
US10476744B2 (en) * 2017-12-15 2019-11-12 Nicira, Inc. Coordinator in cluster membership management protocol
US10572293B2 (en) 2017-12-15 2020-02-25 Nicira, Inc. Node in cluster membership management protocol
JP6714037B2 (en) * 2018-04-26 2020-06-24 株式会社日立製作所 Storage system and cluster configuration control method
US11429441B2 (en) 2019-11-18 2022-08-30 Bank Of America Corporation Workflow simulator
US11106509B2 (en) 2019-11-18 2021-08-31 Bank Of America Corporation Cluster tuner
US11392423B2 (en) * 2019-12-13 2022-07-19 Vmware, Inc. Method for running a quorum-based system by dynamically managing the quorum
US11372553B1 (en) * 2020-12-31 2022-06-28 Seagate Technology Llc System and method to increase data center availability using rack-to-rack storage link cable

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5659777A (en) * 1992-09-25 1997-08-19 Hitachi, Ltd. Method for intraprocessor communication
US5799305A (en) * 1995-11-02 1998-08-25 Informix Software, Inc. Method of commitment in a distributed database transaction

Family Cites Families (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US4925311A (en) * 1986-02-10 1990-05-15 Teradata Corporation Dynamically partitionable parallel processors
JPH02125544A (en) * 1988-11-04 1990-05-14 Mitsubishi Electric Corp Multiple address communication method
IL99923A0 (en) * 1991-10-31 1992-08-18 Ibm Israel Method of operating a computer in a network
JP2615408B2 (en) * 1993-09-20 1997-05-28 工業技術院長 Parallel computer system
EP0678993A1 (en) * 1994-03-30 1995-10-25 International Business Machines Corporation Method and apparatus for controlling the configuration definitions in a data processing system with a plurality of processors
US5666486A (en) * 1995-06-23 1997-09-09 Data General Corporation Multiprocessor cluster membership manager framework
US5918017A (en) * 1996-08-23 1999-06-29 Internatioinal Business Machines Corp. System and method for providing dynamically alterable computer clusters for message routing
US6151688A (en) * 1997-02-21 2000-11-21 Novell, Inc. Resource management in a clustered computer system
JP3790602B2 (en) * 1997-04-25 2006-06-28 富士ゼロックス株式会社 Information sharing device
US6108699A (en) * 1997-06-27 2000-08-22 Sun Microsystems, Inc. System and method for modifying membership in a clustered distributed computer system and updating system configuration
US5999712A (en) * 1997-10-21 1999-12-07 Sun Microsystems, Inc. Determining cluster membership in a distributed computer system

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5659777A (en) * 1992-09-25 1997-08-19 Hitachi, Ltd. Method for intraprocessor communication
US5799305A (en) * 1995-11-02 1998-08-25 Informix Software, Inc. Method of commitment in a distributed database transaction

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
CHANDRA T.D. et al., " On The Impossibility of Group Membership", 1996, ACM 0-89791-800-2/96/05, pages 322-330, XP000829553 *
JOURNAL OF THE ASSOCIATION FOR COMPUTING MACHINERY FISCHER M.J. ET AL: 'IMPOSSIBILITY OF DISTRIBUTED CONSENSUS WITH ONE FAULTY PROSESS' vol. 32, no. 02, April 1985, CONNECTICUT, pages 374 - 382, XP000829554 *

Cited By (18)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
WO2000058825A3 (en) * 1999-03-26 2001-01-04 Microsoft Corp Data distribution in a server cluster
WO2000058825A2 (en) * 1999-03-26 2000-10-05 Microsoft Corporation Data distribution in a server cluster
EP1117039A3 (en) * 2000-01-10 2005-09-07 Sun Microsystems, Inc. Controlled take over of services by remaining nodes of clustered computing system
EP1117041A2 (en) * 2000-01-10 2001-07-18 Sun Microsystems, Inc. Method and apparatus for managing failures in clustered computer systems
EP1117039A2 (en) 2000-01-10 2001-07-18 Sun Microsystems, Inc. Controlled take over of services by remaining nodes of clustered computing system
EP1117041A3 (en) * 2000-01-10 2006-02-22 Sun Microsystems, Inc. Method and apparatus for managing failures in clustered computer systems
EP1134658A2 (en) * 2000-03-14 2001-09-19 Sun Microsystems, Inc. A system and method for comprehensive availability management in a high-availability computer system
US6691244B1 (en) 2000-03-14 2004-02-10 Sun Microsystems, Inc. System and method for comprehensive availability management in a high-availability computer system
EP1134658A3 (en) * 2000-03-14 2002-06-19 Sun Microsystems, Inc. A system and method for comprehensive availability management in a high-availability computer system
EP1877901A2 (en) * 2005-05-06 2008-01-16 Marathon Technologies Corporation Fault tolerant computer system
EP1877901A4 (en) * 2005-05-06 2014-05-07 Stratus Technologies Bermuda Ltd Fault tolerant computer system
JP2015535970A (en) * 2012-09-07 2015-12-17 アビジロン コーポレイション Physical security system having multiple server nodes
US10454997B2 (en) 2012-09-07 2019-10-22 Avigilon Corporation Distributed physical security system
CN110598444A (en) * 2012-09-07 2019-12-20 威智伦公司 Physical security system with multiple server nodes
US9959109B2 (en) 2015-04-10 2018-05-01 Avigilon Corporation Upgrading a physical security system having multiple server nodes
US10474449B2 (en) 2015-04-10 2019-11-12 Avigilon Corporation Upgrading a physical security system having multiple server nodes
US20230229361A1 (en) * 2020-08-14 2023-07-20 Inspur Suzhou Intelligent Technology Co., Ltd. Cluster arbitration method and system based on heterogeneous storage, and device and storage medium
US11762601B2 (en) * 2020-08-14 2023-09-19 Inspur Suzhou Intelligent Technology Co., Ltd. Method for arbitrating heterogeneous storage-based cluster, and system, computer device and non-transitory computer-readable medium thereof

Also Published As

Publication number Publication date
US6449641B1 (en) 2002-09-10
US5999712A (en) 1999-12-07
EP1025506A2 (en) 2000-08-09
CA2306718A1 (en) 1999-04-29
AU1105499A (en) 1999-05-10
WO1999021098A3 (en) 1999-07-01
JP2001521222A (en) 2001-11-06

Similar Documents

Publication Publication Date Title
US5999712A (en) Determining cluster membership in a distributed computer system
EP1402363B1 (en) Method for ensuring operation during node failures and network partitions in a clustered message passing server
US7016946B2 (en) Method and system for establishing a quorum for a geographically distributed cluster of computers
US8621263B2 (en) Automated node fencing integrated within a quorum service of a cluster infrastructure
US5991518A (en) Method and apparatus for split-brain avoidance in a multi-processor system
US6363495B1 (en) Method and apparatus for partition resolution in clustered computer systems
US8140623B2 (en) Non-blocking commit protocol systems and methods
US8850018B2 (en) Consistent cluster operational data in a server cluster using a quorum of replicas
US6889253B2 (en) Cluster resource action in clustered computer system incorporation prepare operation
US6279032B1 (en) Method and system for quorum resource arbitration in a server cluster
US7451359B1 (en) Heartbeat mechanism for cluster systems
US7975006B2 (en) Method and device for managing cluster membership by use of storage area network fabric
US8812501B2 (en) Method or apparatus for selecting a cluster in a group of nodes
JPH11167558A (en) Membership for distributed computer system of low reliability
US20110044209A1 (en) Systems and methods for providing a quiescing protocol
WO1998033121A9 (en) Method and apparatus for split-brain avoidance in a multi-process or system
US7120821B1 (en) Method to revive and reconstitute majority node set clusters
WO2004051474A2 (en) Clustering system and method having interconnect
US7143316B2 (en) Fault diagnosis in a network
EP1704472A1 (en) Configuring a shared resource
JP3446652B2 (en) Hierarchical network management system
Black et al. Determining the Last Membership of a Process Group after a Total Failure

Legal Events

Date Code Title Description
AK Designated states

Kind code of ref document: A2

Designated state(s): AL AM AT AT AU AZ BA BB BG BR BY CA CH CN CU CZ CZ DE DE DK DK EE EE ES FI FI GB GD GE GH GM HR HU ID IL IS JP KE KG KP KR KZ LC LK LR LS LT LU LV MD MG MK MN MW MX NO NZ PL PT RO RU SD SE SG SI SK SK SL TJ TM TR TT UA UG UZ VN YU ZW

AL Designated countries for regional patents

Kind code of ref document: A2

Designated state(s): GH GM KE LS MW SD SZ UG ZW AM AZ BY KG KZ MD RU TJ TM AT BE CH CY DE DK ES FI FR GB GR IE IT LU MC NL PT SE BF BJ CF CG CI CM GA GN GW ML MR NE SN TD TG

DFPE Request for preliminary examination filed prior to expiration of 19th month from priority date (pct application filed before 20040101)
AK Designated states

Kind code of ref document: A3

Designated state(s): AL AM AT AT AU AZ BA BB BG BR BY CA CH CN CU CZ CZ DE DE DK DK EE EE ES FI FI GB GD GE GH GM HR HU ID IL IS JP KE KG KP KR KZ LC LK LR LS LT LU LV MD MG MK MN MW MX NO NZ PL PT RO RU SD SE SG SI SK SK SL TJ TM TR TT UA UG UZ VN YU ZW

AL Designated countries for regional patents

Kind code of ref document: A3

Designated state(s): GH GM KE LS MW SD SZ UG ZW AM AZ BY KG KZ MD RU TJ TM AT BE CH CY DE DK ES FI FR GB GR IE IT LU MC NL PT SE BF BJ CF CG CI CM GA GN GW ML MR NE SN TD TG

121 Ep: the epo has been informed by wipo that ep was designated in this application
ENP Entry into the national phase

Ref document number: 2306718

Country of ref document: CA

Ref country code: CA

Ref document number: 2306718

Kind code of ref document: A

Format of ref document f/p: F

ENP Entry into the national phase

Ref country code: JP

Ref document number: 2000 517348

Kind code of ref document: A

Format of ref document f/p: F

NENP Non-entry into the national phase

Ref country code: KR

WWE Wipo information: entry into national phase

Ref document number: 1998953773

Country of ref document: EP

WWP Wipo information: published in national office

Ref document number: 1998953773

Country of ref document: EP

REG Reference to national code

Ref country code: DE

Ref legal event code: 8642

WWW Wipo information: withdrawn in national office

Ref document number: 1998953773

Country of ref document: EP