System and method for topology-aware job scheduling and backfilling in an HPC environment

• US 8,910,175 B2
• Filed: 10/11/2013
• Issued: 12/09/2014
• Est. Priority Date: 04/15/2004
• Status: Active Grant

• Current Assignee: Raytheon Company (Rtx Corporation)
• Original Assignee: Raytheon Company (Rtx Corporation)
• Inventors: Davidson, Shannon V., Richoux, Anthony N.
• Primary Examiner(s): Vo, Ted T
• Application Number:
  US14/052,114
• Publication Number:
  US 20140047449A1
• Time in Patent Office: 424 Days
• Field of Search: 717159-166, 718/104
• US Class Current: 718/104
• CPC Class Codes: G06F 11/1482 by means of middleware or OS functionality
• CPC Class Codes: G06F 11/2025 using centralised failover control functionality
• CPC Class Codes: G06F 2201/815 Virtual middleware or OS functionality using virtual machines to implement generic software techniques for error detection or fault masking G06F11/1484
• CPC Class Codes: G06F 9/4881 Scheduling strategies for dispatcher, e.g. round robin, multi-level priority queues
• CPC Class Codes: G06F 9/50 Allocation of resources, e.g. of the central processing unit [CPU]
• CPC Class Codes: G06F 9/5038 considering the execution order of a plurality of tasks, e.g. taking priority or time dependency constraints into consideration scheduling strategies G06F9/4881 and subgroups
• CPC Class Codes: G06F 9/5066 Algorithms for mapping a plurality of inter-dependent sub-tasks onto a plurality of physical CPUs mappping at compile time, see G06F8/451
• CPC Class Codes: G06F 9/5072 Grid computing
• CPC Class Codes: G06F 9/5077 Logical partitioning of resources; Management or configuration of virtualized resources specific details on emulation or internal functioning of virtual machines G06F9/455
• CPC Class Codes: G06F 9/5083 Techniques for rebalancing the load in a distributed system

Abstract

A method for job management in an HPC environment includes determining an unallocated subset from a plurality of HPC nodes, with each of the unallocated HPC nodes comprising an integrated fabric. An HPC job is selected from a job queue and executed using at least a portion of the unallocated subset of nodes.

Claims

1. A method comprising:

determining, using one or more computers, available space in a virtual cluster of a plurality of communicatively coupled nodes included in a computing environment, the virtual cluster associated with a group of users that submit similar jobs, and comprising a logical grouping of nodes configured to process related jobs;

determining an optimum job that is compatible with the available space in the virtual cluster of nodes by;

determining a number of available nodes in the virtual cluster;

selecting a first job from a job queue;

dynamically determining an optimum shape of the first job;

determining whether the number of available nodes is enough to execute the first job, based on the optimum shape thereof; and

dynamically allocating one or more of the available nodes for the first job, in the event that the determined number of available nodes is enough to execute the first job;

wherein the optimum shape comprises one or more of;

a best fit cube in which the one or more available nodes are allocated in a cubic volume so as to allow cooperating tasks to exchange data with any other tasks by minimizing the distance between any two nodes; and

a best fit sphere in which the one or more available nodes are allocated in a spherical volume such that a first task is placed in a center node of the sphere with remaining tasks placed on nodes surrounding the center node so as to minimize the distance between the first task and the remaining tasks, wherein the remaining tasks communicate with the first task, but not with each other; and

executing the optimum job in the available space in the virtual cluster of nodes.

2. The method of claim 1, further comprising:

responsive to dynamically allocating one or more of the available nodes for the first job, recalculating the number of available nodes in the virtual cluster; and

executing the first job on the dynamically allocated one or more nodes.

3. The method of claim 2, further comprising:

in the event that either the determined number of available nodes is not enough to execute the first job, or following execution of the first job, selecting a next job in the job queue and repeating the dynamically determining an optimum shape of the next job, determining whether the number of available nodes is enough to execute the next job, based on the optimum shape thereof, and dynamically allocating one or more of the available nodes for the next job, in the event that the determined number of available nodes is enough to execute the next job.

4. The method of claim 3, wherein the job queue is initially sorted based on a job priority.

5. The method of claim 1, wherein the optimum shape is based job dimensions, including one or more of length, width and height.

6. The method of claim 1, wherein the plurality of communicatively coupled nodes included in the computing environment is arranged in a three dimensional node structure.

7. The method of claim 1, wherein determining an optimum job that is compatible with the available space in the virtual cluster of nodes further comprises:

sorting a job queue based on job priority;

determining an original subset of the plurality of communicatively coupled nodes, the original subset comprising nodes currently unallocated to a job;

selecting a first job from the job queue based on priority;

determining that dimensions of the selected job are greater than a topology of the original subset;

in response to the dimensions of the first job being greater than the topology of the original subset, selecting a second job from the sorted job queue; and

executing the selected second job using one or more processors of one or more nodes of the original subset.

8. The method of claim 7, wherein the dimensions of the first job are based in part on one or more job parameters and an associated policy.

9. The method of claim 8, further comprising dynamically allocating a job space from the original subset based on the dimensions of the job, wherein executing the selected job comprises executing the selected job using the dynamically allocated job space.

10. A system comprising:

a plurality of communicatively coupled nodes of a computing environment; and

a management node configured to;

determine available space in a virtual cluster of a plurality of the communicatively coupled nodes, the virtual cluster associated with a group of users that submit similar jobs, and comprising a logical grouping of nodes configured to process related jobs;

determine an optimum job that is compatible with the available space in the virtual cluster of nodes, the optimum job determined by the management node further configured to;

determine a number of available nodes in the virtual cluster;

select a first job from a job queue;

dynamically determine an optimum shape of the first job;

determine whether the number of available nodes is enough to execute the first job, based on the optimum shape thereof; and

dynamically allocate one or more of the available nodes for the first job, in the event that the determined number of available nodes is enough to execute the first job, wherein the optimum shape comprises one or more of;

a best fit cube in which the one or more available nodes are allocated in a cubic volume so as to allow cooperating tasks to exchange data with any other tasks by minimizing the distance between any two nodes; and

a best fit sphere in which the one or more available nodes are allocated in a spherical volume such that a first task is placed in a center node of the sphere with remaining tasks placed on nodes surrounding the center node so as to minimize the distance between the first task and the remaining tasks, wherein the remaining tasks communicate with the first task, but not with each other; and

execute the optimum job in the available space in the virtual cluster of nodes.

11. The system of claim 10, wherein the management node is further configured to:

responsive to dynamically allocating one or more of the available nodes for the first job, recalculate the number of available nodes in the virtual cluster; and

execute the first job on the dynamically allocated one or more nodes.

12. The system of claim 11, wherein the management node is further configured to:

in the event that either the determined number of available nodes is not enough to execute the first job, or following execution of the first job, select a next job in the job queue and repeat the dynamically determining an optimum shape of the next job, determine whether the number of available nodes is enough to execute the next job, based on the optimum shape thereof, and dynamically allocate one or more of the available nodes for the next job, in the event that the determined number of available nodes is enough to execute the next job.

13. The system of claim 12, wherein the job queue is initially sorted based on a job priority.

14. The system of claim 10, wherein the optimum shape is based job dimensions, including one or more of length, width and height.

15. The system of claim 10, wherein the plurality of communicatively coupled nodes included in the computing environment is arranged in a three dimensional node structure.

16. The system of claim 10, wherein the management node is further configured to:

sort a job queue based on job priority;

determine an original subset of the plurality of communicatively coupled nodes, the original subset comprising nodes currently unallocated to a job;

select a first job from the job queue based on priority;

determine that dimensions of the selected job are greater than a topology of the original subset;

in response to the dimensions of the first job being greater than the topology of the original subset, selecting a second job from the sorted job queue; and

execute the selected second job using one or more processors of one or more nodes of the original subset.

17. The system of claim 16, wherein the dimensions of the first job are based in part on one or more job parameters and an associated policy.

18. The system of claim 17, wherein the management node is further configured to dynamically allocate a job space from the original subset based on the dimensions of the job, wherein executing the selected job comprises executing the selected job using the dynamically allocated job space.

19. A non-transitory, computer readable storage medium having computer readable instructions stored thereon that, when executed by a computer, implement a method, the method comprising:

determining available space in a virtual cluster of a plurality of communicatively coupled nodes included in a computing environment, the virtual cluster associated with a group of users that submit similar jobs, and comprising a logical grouping of nodes configured to process related jobs;

determining an optimum job that is compatible with the available space in the virtual cluster of nodes by;

determining a number of available nodes in the virtual cluster;

selecting a first job from a job queue;

dynamically determining an optimum shape of the first job;

determining whether the number of available nodes is enough to execute the first job, based on the optimum shape thereof; and

dynamically allocating one or more of the available nodes for the first job, in the event that the determined number of available nodes is enough to execute the first job;

wherein the optimum shape comprises one or more of;

a best fit cube in which the one or more available nodes are allocated in a cubic volume so as to allow cooperating tasks to exchange data with any other tasks by minimizing the distance between any two nodes; and

a best fit sphere in which the one or more available nodes are allocated in a spherical volume such that a first task is placed in a center node of the sphere with remaining tasks placed on nodes surrounding the center node so as to minimize the distance between the first task and the remaining tasks, wherein the remaining tasks communicate with the first task, but not with each other; and

executing the optimum job in the available space in the virtual cluster of nodes.

20. The storage medium of claim 19, wherein the method further comprises:

responsive to dynamically allocating one or more of the available nodes for the first job, recalculating the number of available nodes in the virtual cluster; and

executing the first job on the dynamically allocated one or more nodes.

21. The storage medium of claim 20, wherein the method further comprises:

in the event that either the determined number of available nodes is not enough to execute the first job, or following execution of the first job, selecting a next job in the job queue and repeating the dynamically determining an optimum shape of the next job, determining whether the number of available nodes is enough to execute the next job, based on the optimum shape thereof, and dynamically allocating one or more of the available nodes for the next job, in the event that the determined number of available nodes is enough to execute the next job.

22. The storage medium of claim 21, wherein the job queue is initially sorted based on a job priority.

23. The storage medium of claim 19, wherein the optimum shape is based job dimensions, including one or more of length, width and height.

24. The storage medium of claim 19, wherein the plurality of communicatively coupled nodes included in the computing environment is arranged in a three dimensional node structure.

25. The storage medium of claim 19, wherein determining an optimum job that is compatible with the available space in the virtual cluster of nodes further comprises:

sorting a job queue based on job priority;

determining an original subset of the plurality of communicatively coupled nodes, the original subset comprising nodes currently unallocated to a job;

selecting a first job from the job queue based on priority;

determining that dimensions of the selected job are greater than a topology of the original subset;

in response to the dimensions of the first job being greater than the topology of the original subset, selecting a second job from the sorted job queue; and

executing the selected second job using one or more processors of one or more nodes of the original subset.

26. The storage medium of claim 25, wherein the dimensions of the first job are based in part on one or more job parameters and an associated policy.

27. The storage medium of claim 26, wherein the method further comprises dynamically allocating a job space from the original subset based on the dimensions of the job, wherein executing the selected job comprises executing the selected job using the dynamically allocated job space.