CN1315046C - A method for allocating computation nodes in cluster job management system - Google Patents
A method for allocating computation nodes in cluster job management system Download PDFInfo
- Publication number
- CN1315046C CN1315046C CNB2004100294247A CN200410029424A CN1315046C CN 1315046 C CN1315046 C CN 1315046C CN B2004100294247 A CNB2004100294247 A CN B2004100294247A CN 200410029424 A CN200410029424 A CN 200410029424A CN 1315046 C CN1315046 C CN 1315046C
- Authority
- CN
- China
- Prior art keywords
- node
- load balance
- balance index
- average process
- calculating
- Prior art date
- Legal status (The legal status 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 status listed.)
- Expired - Fee Related
Links
Abstract
The present invention discloses a method for allocating computation nodes in a cluster job management system. The method comprises the steps that load balance indexes of a cluster system are determined; the value of each load balance index is determined according to user jobs; the load balance exponent of each computation node is calculated; the computation nodes of the user jobs are selected to be executed according to the load balance exponent of each computation node. The computation nodes allocated by the method are very accordant with the requirements on computation node resources by the user jobs; the computation nodes are reasonably assigned to the user jobs, the computational efficiency of the user jobs is greatly improved; besides, the load among the computation nodes is balanced, and the resource utilization rate and the throughput rate of the cluster system are markedly raised.
Description
Technical field
The present invention relates to the cluster job management system technology, the method for Distribution Calculation node in particularly a kind of cluster job management system.
Background technology
Cluster job management system is used for the operation and the resource of unified management and scheduling Network of Workstation.The core content of cluster job management system is a job scheduling, and job scheduling comprises usually to the choosing method of user job with to calculating the distribution method of node.Choosing of user job mainly contained first in first out (FIFO), first be fit to several different methods such as (FirstFit), the most suitable (BestFit), reservation (Reservation) and filling (Backfill) combine, and it is relatively single to calculate the distribution method of node at present.
In cluster job management system, at present the method for Distribution Calculation node commonly used be to determine that at first each calculates node interior average process number of the schedule time in the past, and then in conjunction with the concrete Distribution Calculation node of number of the specified calculating node of user job.
Figure 1 shows that the process flow diagram of the Distribution Calculation node method of prior art.As shown in Figure 1, may further comprise the steps:
Step 101: determine respectively to calculate node interior average process number of the schedule time in the past;
Step 102:, choose the less calculating node of average process number successively according to calculating node number specified in the user job.
In actual application, user job has nothing in common with each other to the demand of resources such as CPU, internal memory, page or leaf swapace and disk.For example: certain user's operation may need to take a large amount of CPU time when operation, and other user jobs may need to take bigger memory headroom and space, exchange area when operation.Simultaneously, take the user job of CPU time in a large number, also should consider the stability of CPU during the Distribution Calculation node, so cluster job management system need be given user job Distribution Calculation node according to the actual demand of user job for needs.
The calculating node distribution method of prior art has only considered to calculate the node average number of processes in certain period in the past when the Distribution Calculation node, do not take all factors into consideration the operating position of the resources such as internal memory, page or leaf swapace, soft ware authorization number and disk of calculating node flexibly according to the concrete needs of user job operation, also consider the stability of the CPU of calculating node simultaneously, therefore can not be effectively match according to prior art institute distribution calculation node, thereby reduced the counting yield of user job with the resource requirement of user job.
Simultaneously, Distribution Calculation node not according to the actual needs of user job and reasonably in the prior art causes each to calculate laod unbalance between node, thereby has reduced the resource utilization and the throughput of Network of Workstation.
Summary of the invention
In view of this, fundamental purpose of the present invention is the method that proposes Distribution Calculation node in a kind of cluster job management system, to improve the counting yield of user job.
For achieving the above object, technical scheme of the present invention is such:
The method of Distribution Calculation node in a kind of cluster job management system may further comprise the steps:
A, definite load balance index of calculating node, described load balance index comprise at least calculates node interior average process number of the schedule time in the past;
B, determine the weights of described each load balance index according to user job, and the described schedule time in past was set to more than 1 time period, when for the average process number in the described schedule time in the past weights being set, and the average process number that is respectively in described each time period is provided with weights;
C, calculate the load balance index of assignable calculating node, calculate described average process of passing by in the schedule time according to the average process numerical value of each time period and the respective weights of average process number in this time period and count numerical value according to the weights of the numerical value of the described load balance index of calculating node and described each load balance index of step B;
D, select to carry out the calculating node of user job according to the described load balance index of step C.
The described load balance index of steps A further comprises or combination in any in cpu busy percentage, memory usage, disk utilization rate, soft ware authorization number, the page or leaf exchange frequency.
The method of the load balance index of the assignable calculating node of the described calculating of step C is: calculate the product of the weight of the numerical value of each load balance index of assignable calculating node and this load balance index, then with described product summation.
Average process number in the described calculating of the step C schedule time in the past is: calculate the average process numerical value of each time period and the product of this time period respective weights, then with described product summation.
Step B is described to determine that the method for the weights of each load balance index is: according to demand from high to low the order of user job to load balance index, for load balance index is provided with descending weights, the method that the calculating node of user job is carried out in the described selection of step D is: select to calculate node successively according to load balance index order from small to large.
According to demand from high to low the order of user job to load balance index, for load balance index is provided with ascending weights, the method that the calculating node of user job is carried out in the described selection of step D is: select to calculate node successively according to load balance index order from big to small.
Further before step C, preestablish average process and count max-thresholds, and judge whether the average process number that calculates node surpasses this predefined average process and count max-thresholds, if surpassing then this calculating node is set is indistributable calculating node, execution in step C then is if be no more than then direct execution in step C.
Further before step C, preestablish average process and count desired threshold, when the average process number that calculates node is counted max-thresholds and is reduced to and is less than or equal to average process when counting max-thresholds from surpassing average process, whether the average process number of judging this calculating node is lower than described predefined average process is counted desired threshold, if be lower than then direct execution in step C, if being not less than then this calculating node is set is indistributable calculating node, execution in step C then.
As can be seen from the above technical solutions, the present invention at first determines load balance index according to the resource situation of cluster job management system, real needs according to user job are provided with corresponding weights for these load balance index then, then the weights according to load balance index and the load balance index of determining calculate the load balance index that each calculates node, determine the calculating node of execution user job at last according to specified calculating nodal point number in the load balance exponential sum user job that calculates node.So the present invention is when being user job Distribution Calculation node, not only considered to calculate the interior average process number of the schedule time in the past of node, also taken all factors into consideration internal memory, the page or leaf swapace, the operating position of resources such as disk, therefore the present invention has realized coming the Distribution Calculation node according to the behaviour in service of calculating every resource on the node and the real needs of user job, so institute's distribution calculation node and user job are very identical to the demand of resource behind application the present invention, more reasonably for user job has distributed the calculating node, thereby greatly improved the counting yield of user job.
Simultaneously, the present invention calculates the utilization of resources situation of node in conjunction with each and concrete user job demand assignment is calculated node, and count the calculating node of max-thresholds and average process number for average process number greater than average process and count the operation of not reallocating of max-thresholds descends but still count desired threshold greater than average process calculating node from average process, load between the feasible calculating node is balance more, thereby has greatly improved the resource utilization and the throughput of Network of Workstation.
Description of drawings
Fig. 1 is a prior art Distribution Calculation node schematic flow sheet.
Fig. 2 is one embodiment of the invention Distribution Calculation node schematic flow sheet.
Embodiment
For making the purpose, technical solutions and advantages of the present invention express clearlyer, the present invention is further described in more detail below in conjunction with drawings and the specific embodiments.
Main thought of the present invention is to come the Distribution Calculation node according to the resource situation of calculating node and the real needs of user job.At first determine every load balance index according to the resource situation of calculating node, real needs according to user job are provided with corresponding weights for these load balance index then, then the weights according to load balance index and the load balance index of determining calculate the load balance index that each calculates node, select to carry out the calculating node of user job at last according to the load balance index.
Fig. 2 is the Distribution Calculation node schematic flow sheet of one embodiment of the invention.As shown in Figure 2, may further comprise the steps:
Step 201: determine machine group system load balance index.
For the load balancing of the calculating node of realizing Network of Workstation, when selecting to calculate node, need take all factors into consideration the loading condition of Network of Workstation, the factor that can influence the Network of Workstation load is set to corresponding load balance index.The factor that influences the Network of Workstation load has a lot, for example cpu busy percentage, page or leaf exchange frequency, memory usage, soft ware authorization number and disk utilization rate etc., also must consider the CPU behaviour in service simultaneously, and the behaviour in service of CPU can embody with the average process number in the schedule time in past, so cluster job management system can be set to load balance index with the average process number in schedule time in the past, cpu busy percentage, memory usage, soft ware authorization number, disk utilization rate and page or leaf exchange frequency etc.
Step 202:, set the weights of each load balance index according to the needs of user job.
Different user jobs is different to the demand of calculating the every resource of node, therefore can set corresponding weights for each load balance index according to the actual needs of user job.For example: when user job need take a certain resource significantly, can higher weights be set with the pairing load balance index of this resource.When user job is not high to the demand of a certain resource, can give the load balance index setting lower weights corresponding with this resource.For example, if user job need take a large amount of internal memories, the expression user job requires the memory usage of calculating node very low, higher weights is set then can for this load balance index of memory usage; If user job need take less disk space, the expression user job allows the disk utilization rate of calculating node higher, then can be this load balance index of disk utilization rate less weights are set.
In like manner, when user job need take a certain resource significantly, also can give and the lower weights of the pairing load balance index setting of this resource.At this moment, in the step 204 of back, select the bigger calculating node of load balance index successively according to the specified calculating nodal point number of user job.
Step 203:, calculate the load balance index that respectively calculates node according to the numerical value of the load balance index of calculating node and the weights that calculate the node load balance index.
After the corresponding weight value of determining system load balance index and each system load balance index, can adopt certain load balancing algorithm to come to calculate in the computer cluster system load balance index of node.For example the weights of the concrete numerical value of a certain load balance index and this load balance index can be multiplied each other weights sum divided by all load balance index to obtain the load factor of this resource again, and then to the summation of the load factor of every resource, resulting and be exactly the load balance index that calculates node.Be that the load balance index is the sum of products of the weight of every load balance index numerical value and correspondence.
Suppose certain calculate node in the past average process number and the weights thereof in the certain hour be respectively a1 and b1, the utilization factor of CPU and weights thereof are respectively a2 and b2, memory usage and weights thereof are respectively a3 and b3, page or leaf exchange frequency and weights thereof are respectively a4 and b4, disk utilization rate and weights thereof are respectively a5 and b5, the load balance index of this calculating node is Y, and the load balance index is the sum of products of the weight of every load balance index numerical value and correspondence so.So:
Y＝(a1×b1)/(b1+b2+b3+b4+b5)+(a2×b2)/(b1+b2+b3+b4+b5)+(a3×b3)/(b1+b2+b3+b4+b5)+(a4×b4)/(b1+b2+b3+b4+b5)+(a5×b5)/(b1+b2+b3+b4+b5)
In order to make selected calculating node be fit to the demand of user job more, when the average process in definite schedule time is counted, can also further be divided into several time periods to this section schedule time, and set corresponding weights and then calculating again for the average process number in each time period after further dividing, wherein when user job need calculate node advantages of higher stability is arranged, can be with the higher weights that are provided with of the average process number in the big time period, thus represent to calculate in the certain hour behaviour in service of node more accurately.
If for example choose with in the past 15 minutes and represent that the average process of calculating node counts situation, to the 10th minute in the past the average process number and the average process number in past the 1st minute to past the 15th minute weights were set can be respectively for the 1st minute in the past to the 5th minute in the past average process number, past the 1st minute, and then in conjunction with the weights calculating of the average process numbers of this three time periods load factor of the average process number in 15 minutes in the past.On average process number and weights thereof are respectively c1 and d1 in the 1st minute to the 5th minute in the past in the past to suppose to calculate node, average process number and weights thereof in the past the 1st minute to the 10th minute in the past are respectively c2 and d2, average process and weights thereof in the past the 1st minute to the 15th minute in the past are respectively c3 and d3, the load factor of supposing the average process number in 15 minutes of this past is A, the weights of the load factor of the average process number in past 15 minutes are b1, so:
A＝(c1×d1)/(d1+d2+d3)+(c2×d2)/(d1+d2+d3)+(c3×d3)/(d1+d2+d3)
At this moment, load balance index Y is:
Y＝(A×b1)/(b1+b2+b3+b4+b5)+(a2×b2)/(b1+b2+b3+b4+b5)+(a3×b3)/(b1+b2+b3+b4+b5)+(a4×b4)/(b1+b2+b3+b4+b5)+(a5×b5)/(b1+b2+b3+b4+b5)
After calculating each load balance index that calculates node, just can choose the calculating node successively according to the load balance index.
Step 204:, choose the specified calculating node number of user job successively according to the order of load balance index.
Step 202 is when setting the weights of each load balance index, when user job need take a certain resource significantly, if be provided with higher weights for the pairing load balance index of this resource, when selecting to calculate node, should select the less calculating node of load balance index to carry out user job successively according to the specified calculating nodal point number of user job.Correspondingly, if when step 202 is provided with weights, be provided with higher weights when user job less needs to take a certain resource, for the load balance index corresponding, select the bigger calculating node of load balance index in step 204 successively according to the specified calculating nodal point number of user job so with this resource.
In the above process, can preestablish the average process of calculating node and count max-thresholds, take a step forward in step 203 and to judge that whether average process number surpasses average process and count max-thresholds, if surpass, calculating node state so will be masked as and can not distribute and do not calculate the load balance index of this calculating node, and no longer receive user job, if be no more than then this calculating node be set for distributing, and execution in step 203.
When the average process that preestablishes the calculating node is counted max-thresholds, also can further preestablish the average process of calculating node and count desired threshold.When the average process number that calculates node drops to when counting max-thresholds less than average process by count max-thresholds more than or equal to average process, judge further whether average process number counts desired threshold greater than average process, if this moment, average process number was still greater than the desired load threshold value, then this calculates node state and still is masked as and can not distributes and do not calculate the load balance index of this calculating node, do not receive user job, to prevent the transition of average number of processes.Have only the average process number that calculates node to count desired threshold or average process number and count desired threshold from average process and increase but be no more than average process and count the calculating node of max-thresholds and just be set to and can distribute less than average process, and execution in step 203 and later step thereof.
In the above process, if, think that then user job does not have specific (special) requirements to every resource, so cluster job management system can provide identical default weights at step 202 weights of appointing system balance index not.
Describe with a concrete example below.Be provided with a Network of Workstation, wherein cluster job management system is used for unified management and scheduling Network of Workstation operation and resource, and this cluster job management system need be given certain user job Distribution Calculation node.Suppose that assignable calculating nodal point number is 3 in this Network of Workstation, be respectively and calculate node a, calculate node b, calculate node c.Wherein calculate node a in the past the average process number in 15 minutes be 0.6, memory usage is 0.82, the disk utilization rate is 0.27; The calculating node b average process number in 15 minutes in the past is 0.4, and memory usage is 0.98, and the disk utilization rate is 0.23; The calculating node c average process number in 15 minutes in the past is 0.0, and memory usage is 0.85, and the disk utilization rate is 0.23.The calculating node number of supposing appointment in this user job is 2, and require to carry out that the average process number of calculating node of this user job is considerably less, memory usage is lower, and this user job less demanding to the disk utilization rate.
As shown in Figure 2, have the following steps successively:
In step 201:, choose load balance index according to the resource situation of Network of Workstation.Load balance index is for calculating average process number, memory usage and the disk utilization rate in the node 15 minutes herein.
In step 202:, the weights of load balance index are set according to the actual needs of user job to resource.The average process number of calculating node is considerably less, memory usage is lower because user job requires, requirement is not very high and to the disk utilization rate, so give the very little weights of disk utilization rate setting, less weights be set and higher weights be set for average process number to memory usage.The weights of the average process number in 15 minutes of can setting over are 30, and the weights of memory usage are 20, and the weights of disk utilization rate are 10.
In step 203: the load balance index of determining respectively to calculate node.
The load balance index that calculates node a is:
(0.6×30)/(30+20+10)+(0.82×20)/(30+20+10)+(0.27×10)/(30+20+10)＝0.618
The load balance index that calculates node b is:
(0.4×30)/(30+20+10)+(0.98×20)/(30+20+10)+(0.23×10)/(30+20+10)＝0.565
The load balance index that calculates node c is:
(0.0×30)/(30+20+10)+(0.85×20)/(30+20+10)+(0.23×10)/(30+20+10)＝0.322
In step 204: according to the load balance index Distribution Calculation node that calculates node.Relatively calculate the load balance index of node a, calculating node b and calculating node c, choose the less calculating node of load balance index successively, so select to calculate node c and calculating node b execution user job.
In the above example, can preestablish the average process of calculating node and count max-thresholds, be assumed to be 2.0.Suppose that this Network of Workstation also has one to calculate node d except calculating node a, calculating node b and calculate the node c, and calculate node d in the past the average number of processes in 15 minutes be 2.3.Because the calculating node d average number process in 15 minutes in the past counts max-thresholds greater than average process, so when giving user job Distribution Calculation node, then will calculate node d and foreclose.Simultaneously, can also further preestablish the average process of calculating node and count desired threshold, be assumed to be 1.5.When calculating node d the average number in 15 minutes be reduced to 1.6 by 2.3 in the past, though calculate this moment node d in the past the average process number in 15 minutes count max-thresholds 2.0 less than average process, but owing to still count desired threshold 1.5 greater than average process, so still do not consider this moment to count transition to prevent average process to calculating the operation of node d distributing user.Have only when the average process number that calculates node d is counted desired threshold 0.6 less than average process, just will calculate node d and participate in selecting.
The above is preferred embodiment of the present invention only, is not to be used to limit protection scope of the present invention.Within the spirit and principles in the present invention all, any modification of being done, be equal to replacement, improvement etc., all should be included within protection scope of the present invention.
Claims (8)
1, the method for Distribution Calculation node in a kind of cluster job management system is characterized in that, may further comprise the steps:
A, definite load balance index of calculating node, described load balance index comprise at least calculates node interior average process number of the schedule time in the past;
B, determine the weights of described each load balance index according to user job, and the described schedule time in past was set to more than 1 time period, when for the average process number in the described schedule time in the past weights being set, and the average process number that is respectively in described each time period is provided with weights;
C, calculate the load balance index of assignable calculating node, calculate described average process of passing by in the schedule time according to the average process numerical value of each time period and the respective weights of average process number in this time period and count numerical value according to the weights of the numerical value of the described load balance index of calculating node and described each load balance index of step B;
D, select to carry out the calculating node of user job according to the described load balance index of step C.
2, the method for Distribution Calculation node in the cluster job management system according to claim 1, it is characterized in that the described load balance index of steps A further comprises or combination in any in cpu busy percentage, memory usage, disk utilization rate, soft ware authorization number, the page or leaf exchange frequency.
3, the method for Distribution Calculation node in the cluster job management system according to claim 1, it is characterized in that, the method of the load balance index of the assignable calculating node of the described calculating of step C is: calculate the product of the weight of the numerical value of each load balance index of assignable calculating node and this load balance index, then with described product summation.
4, the method for Distribution Calculation node in the cluster job management system according to claim 1, it is characterized in that, average process number in the described calculating of the step C schedule time in the past is: calculate the average process numerical value of each time period and the product of this time period respective weights, then with described product summation.
5, the method for Distribution Calculation node in the cluster job management system according to claim 1, it is characterized in that, step B is described to determine that the method for the weights of each load balance index is: according to demand from high to low the order of user job to load balance index, for load balance index is provided with descending weights, the method that the calculating node of user job is carried out in the described selection of step D is: select to calculate node successively according to load balance index order from small to large.
6, the method for Distribution Calculation node in the cluster job management system according to claim 1, it is characterized in that, according to demand from high to low the order of user job to load balance index, for load balance index is provided with ascending weights, the method that the calculating node of user job is carried out in the described selection of step D is: select to calculate node successively according to load balance index order from big to small.
7, the method for Distribution Calculation node in the cluster job management system according to claim 1, it is characterized in that, further before step C, preestablish average process and count max-thresholds, and judge whether the average process number that calculates node surpasses this predefined average process and count max-thresholds, if surpassing then this calculating node is set is indistributable calculating node, execution in step C then is if be no more than then direct execution in step C.
8, the method of Distribution Calculation node in the cluster job management system according to claim 7, it is characterized in that, further before step C, preestablish average process and count desired threshold, when the average process number that calculates node is counted max-thresholds and is reduced to and is less than or equal to average process when counting max-thresholds from surpassing average process, whether the average process number of judging this calculating node is lower than described predefined average process is counted desired threshold, if be lower than then direct execution in step C, if being not less than then this calculating node is set is indistributable calculating node, execution in step C then.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CNB2004100294247A CN1315046C (en) | 2004-03-17 | 2004-03-17 | A method for allocating computation nodes in cluster job management system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CNB2004100294247A CN1315046C (en) | 2004-03-17 | 2004-03-17 | A method for allocating computation nodes in cluster job management system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN1670706A CN1670706A (en) | 2005-09-21 |
| CN1315046C true CN1315046C (en) | 2007-05-09 |
Family
ID=35041979
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CNB2004100294247A Expired - Fee Related CN1315046C (en) | 2004-03-17 | 2004-03-17 | A method for allocating computation nodes in cluster job management system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN1315046C (en) |
Families Citing this family (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN100451970C (en) * | 2005-09-27 | 2009-01-14 | 浪潮电子信息产业股份有限公司 | Method for realizing dynamic layout of high-performance server based on group structure |
| CN101667034B (en) * | 2009-09-21 | 2011-06-22 | 北京航空航天大学 | Scalable monitoring system supporting hybrid clusters |
| CN102193832B (en) * | 2010-03-11 | 2013-03-27 | 精英电脑股份有限公司 | Cloud computing resource scheduling method and applied system |
| CN103064741B (en) * | 2012-12-24 | 2015-08-05 | 浙江工业大学 | A kind of method of the divided load scheduling based on energy model |
| CN103401947A (en) * | 2013-08-20 | 2013-11-20 | 曙光信息产业(北京)有限公司 | Method and device for allocating tasks to multiple servers |
| CN108153584B (en) * | 2016-12-02 | 2021-11-09 | 百度在线网络技术（北京）有限公司 | Method and equipment for determining number of threads to be allocated to target network equipment |
| CN109271251A (en) * | 2018-08-09 | 2019-01-25 | 深圳市瑞云科技有限公司 | A method of by constraining come scheduling node machine |
| CN109800087A (en) * | 2019-01-02 | 2019-05-24 | 郑州云海信息技术有限公司 | A method of for automatic test |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6067545A (en) * | 1997-08-01 | 2000-05-23 | Hewlett-Packard Company | Resource rebalancing in networked computer systems |
| US20020078263A1 (en) * | 2000-12-18 | 2002-06-20 | Darling Christopher L. | Dynamic monitor and controller of availability of a load-balancing cluster |
| CN1460656A (en) * | 2003-06-27 | 2003-12-10 | 吴钢 | Colorant for producing golden flower rice yellow ceramic polished brick |
-
2004
- 2004-03-17 CN CNB2004100294247A patent/CN1315046C/en not_active Expired - Fee Related
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US6067545A (en) * | 1997-08-01 | 2000-05-23 | Hewlett-Packard Company | Resource rebalancing in networked computer systems |
| US20020078263A1 (en) * | 2000-12-18 | 2002-06-20 | Darling Christopher L. | Dynamic monitor and controller of availability of a load-balancing cluster |
| CN1460656A (en) * | 2003-06-27 | 2003-12-10 | 吴钢 | Colorant for producing golden flower rice yellow ceramic polished brick |
Also Published As
| Publication number | Publication date |
|---|---|
| CN1670706A (en) | 2005-09-21 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN100343810C (en) | Task Scheduling method, system and apparatus | |
| CN1681263A (en) | Method and system for actively managing central queue buffer allocation using a backpressure mechanism | |
| US8286170B2 (en) | System and method for processor thread allocation using delay-costs | |
| CN106851747B (en) | Dynamic resource allocation method and device in mobile communication system | |
| CN102096599A (en) | Multi-queue task scheduling method and related system and equipment | |
| CN1315046C (en) | A method for allocating computation nodes in cluster job management system | |
| CN105718316A (en) | Job scheduling method and apparatus | |
| CN114996001A (en) | Distributed machine learning task GPU resource scheduling and distributing method and system | |
| CN101694630A (en) | Method, system and equipment for operation dispatching | |
| CN113568721A (en) | Task scheduling method and related equipment | |
| CN1643874A (en) | Scheduling using quantum and deficit values | |
| CN105320565A (en) | Computer resource scheduling method for various application software | |
| CN1553693A (en) | Universal charging method | |
| CN1889521A (en) | Data dispatching method and system | |
| CN1665315A (en) | Method for controlling overload of intelligent network based on service control point in multi-service environment | |
| CN116483538A (en) | Data center task scheduling method with low consistency and delay | |
| CN1302412C (en) | Computer group system and its operation managing method | |
| CN114860449B (en) | Data processing method, device, equipment and storage medium | |
| CN1798087A (en) | Method of shaping network flow in use for multiple inputs and single output, and of distributing bandwidth | |
| CN111628943B (en) | Intelligent Internet of things method based on communication and perception technology | |
| CN1773962A (en) | Gigabit passive optical network strict priority weighted round robin scheduling mechanism | |
| CN1302644C (en) | Method for building permanent virtual connection | |
| CN1521988A (en) | Method for uniform distribution of physical layer data transmitting time slot in whole time domain | |
| CN1863082A (en) | Method of dynamic allocation of network communication apparatus system resource | |
| CN111324447A (en) | Thread adjusting method and device, application server and storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20070509 Termination date: 20210317 |