Approach based on self-evolving models for performance guarantees in a shared storage system

• US 20070112723A1
• Filed: 11/16/2005
• Published: 05/17/2007
• Est. Priority Date: 11/16/2005
• Status: Active Grant

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Alvarez, Guillermo, Palmer, John, Yin, Li, Uttamchandani, Sandeep
• Granted Patent: US 7,640,231 B2
• Time in Patent Office: Days
• US Class Current: 1/1
• CPC Class Codes: G06F 9/50 Allocation of resources, e.g. of the central processing unit [CPU]
• CPC Class Codes: G06F 9/5083 Techniques for rebalancing the load in a distributed system

Abstract

A technique of allocating shared resources in a computer network-based storage system comprises taking periodic performance samples on a running storage system; evaluating an objective function that takes as input the performance samples to quantify how aligned a current state of the storage system is with organizational objectives; building and maintaining models of behavior and capabilities of the storage system by using the performance samples as input; determining how resources of the storage system should be allocated among client computers in the storage system by selecting one among many possible allocations based on predictions generated by the models in order to maximize a value of the objective function; calculating a confidence statistic value for a chosen resource allocation based on an accuracy of the models; and enforcing the chosen resource allocation on the running storage system when the confidence statistic value is at or above a predetermined threshold value.

Claims

1. A computer-implemented method of allocating shared resources in a computer network-based storage system, said method comprising:

taking periodic performance samples on a running computer network-based storage system;

evaluating an objective function that takes as input said performance samples to quantify how aligned a current state of said computer network-based storage system is with organizational objectives;

building and maintaining models of behavior and capabilities of said computer network-based storage system by using said performance samples as input;

determining how resources of said computer network-based storage system should be allocated among client computers in said computer network-based storage system by selecting one among many possible allocations based on predictions generated by said models in order to maximize a value of said objective function;

calculating a confidence statistic value for a chosen resource allocation based on an accuracy of said models; and

enforcing said chosen resource allocation on the running computer network-based storage system when said confidence statistic value is at or above a predetermined threshold value.

2. The method of claim 1, further comprising:

maintaining a predetermined set of system designer-defined policies for allocating said resources among said client computers; and

using said set of system designer-defined policies to allocate said resources among said client computers so long as said confidence statistic value fails to reach said predetermined threshold value.

3. The method of claim 1, wherein a Service Level Agreement (SLA) is associated to at least partially some of said client computers that specifies a performance that each associated client computer expects to experience in exchange for respecting specified bounds on a load said client computer imposes on said computer network-based storage system, and wherein a value of said objective function depends on which SLAs are being violated.

4. The method of claim 3, wherein at least one client computer with an associated SLA is assigned a priority such that said objective function incorporates said priority to prefer system states in which said SLAs of said client computers with higher priorities are satisfied.

5. The method of claim 3, wherein said models comprise data relating to:

capabilities of individual components in said computer network-based storage system;

characteristics of a workload presented to said components; and

an expected response from the workloads to changes in the amount of system resources available to said workloads.

6. The method of claim 5, wherein said performance samples are used to improve an accuracy of said models in a periodic and fully-automatic manner, and to determine when variable system conditions have caused a discontinuous change in the behavior of said computer network-based storage system.

7. The method of claim 6, wherein incremental model refinement is accomplished starting from any of:

a model built by taking performance samples from said computer network-based storage system in a non-production use;

a model built by taking performance samples from said computer network-based storage system in a production use while models are not being used to allocate resources; and

a simplified representation of a behavior of said computer network-based storage system.

8. The method of claim 5, wherein a determination of an allocation of system resources among competing client computers is made by solving a constrained optimization problem.

9. The method of claim 8, further comprising allocating additional resources to at least one particular client computer when a SLA requirement of said particular client computer is being insufficiently met.

10. The method of claim 9, wherein said allocating of additional resources to said particular client computer occurs by reducing an amount of resources provided to another one of said client computers, and reallocating said resources to said particular client computer.

11. The method of claim 10, wherein said reducing an amount of resources provided to another one of said client computers occurs by throttling workloads of competing client computers.

12. The method of claim 11, further comprising computing a rate at which each workload stream for said client computers should be allowed to issue inputs/outputs (I/Os) to said computer network-based storage system.

13. The method of claim 12, further comprising:

for each component used by said particular client computer having an underperforming workload, using said model to determine a change in a request rate at said component of a storage component in said computer network-based storage system required to make said underperforming workload reach a sufficient level of performance;

querying said model for each workload using said component to determine which change in an input/output (I/O) injection rate of said workload is necessary to relieve a load of said component; and

determining a token issue rate necessary for a desired change in said injection rate.

14. The method of claim 13, wherein transitions between current workload token issue rates and lower token issue rates are enforced progressively in steps whose size depends on said confidence statistical value.

15. The method of claim 14, further comprising reassessing token issue rates depending on a change in said load of said component.

16. The method of claim 15, further comprising analyzing a current state of performance of said computer network system after a predetermined number of throttling steps.

17. The method of claim 16, wherein token issue rates for one or more said client computers are increased by unthrottling their workloads when system resources become available.

18. A program storage device readable by a computer, tangibly embodying a program of instructions executable by said computer to perform computer-implemented method of allocating shared resources in a computer network-based storage system, said method comprising:

taking periodic performance samples on a running computer network-based storage system;

evaluating an objective function that takes as input said performance samples to quantify how aligned a current state of said computer network-based storage system is with organizational objectives;

building and maintaining models of behavior and capabilities of said computer network-based storage system by using said performance samples as input;

determining how resources of said computer network-based storage system should be allocated among client computers in said computer network-based storage system by selecting one among many possible allocations based on predictions generated by said models in order to maximize a value of said objective function;

calculating a confidence statistic value for a chosen resource allocation based on an accuracy of said models; and

enforcing said chosen resource allocation on the running computer network-based storage system when said confidence statistic value is at or above a predetermined threshold value.

19. The program storage device of claim 18, wherein said method further comprises:

maintaining a predetermined set of system designer-defined policies for allocating said resources among said client computers; and

using said set of system designer-defined policies to allocate said resources among said client computers so long as said confidence statistic value fails to reach said predetermined threshold value.

20. The program storage device of claim 18, wherein a Service Level Agreement (SLA) is associated to at least partially some of said client computers that specifies a performance that each associated client computer expects to experience in exchange for respecting specified bounds on a load said client computer imposes on said computer network-based storage system, and wherein a value of said objective function depends on which SLAs are being violated.

21. The program storage device of claim 20, wherein at least one client computer with an associated SLA is assigned a priority such that said objective function incorporates said priority to prefer system states in which said SLAs of said client computers with higher priorities are satisfied.

22. The program storage device of claim 20, wherein said models comprise data relating to:

capabilities of individual components in said computer network-based storage system;

characteristics of a workload presented to said components; and

an expected response from the workloads to changes in the amount of system resources available to said workloads.

23. The program storage device of claim 22, wherein said performance samples are used to improve an accuracy of said models in a periodic and fully-automatic manner, and to determine when variable system conditions have caused a discontinuous change in the behavior of said computer network-based storage system.

24. The program storage device of claim 23, wherein incremental model refinement is accomplished using any of:

a model built by taking performance samples from said computer network-based storage system in a non-production use;

a model built by taking performance samples from said computer network-based storage system in a production use while models are not being used to allocate resources; and

a simplified representation of a behavior of said computer network-based storage system.

25. The program storage device of claim 22, wherein a determination of an allocation of system resources among competing client computers is made by solving a constrained optimization problem.

26. The program storage device of claim 25, wherein said method further comprises allocating additional resources to at least one particular client computer when a SLA requirement of said particular client computer is being insufficiently met.

27. The program storage device of claim 26, wherein said allocating of additional resources to said particular client computer occurs by reducing an amount of resources provided to another one of said client computers, and reallocating said resources to said particular client computer.

28. The program storage device of claim 27, wherein said reducing an amount of resources provided to another one of said client computers occurs by throttling workloads of competing client computers.

29. The program storage device of claim 28, wherein said method further comprises computing a rate at which each workload stream for said client computers should be allowed to issue inputs/outputs (I/Os) to said computer network-based storage system.

30. The program storage device of claim 29, wherein said method further comprises:

for each component used by said particular client computer having an underperforming workload, using said model to determine a change in a request rate at said component of a storage component in said computer network-based storage system required to make said underperforming workload reach a sufficient level of performance;

querying said model for each workload using said component to determine which change in an input/output (I/O) injection rate of said workload is necessary to relieve a load of said component; and

determining a token issue rate necessary for a desired change in said injection rate.

31. The program storage device of claim 30, wherein transitions between current workload token issue rates and lower token issue rates are enforced progressively in steps whose size depends on said confidence statistical value.

32. The program storage device of claim 31, wherein said method further comprises reassessing token issue rates depending on a change in said load of said component.

33. The program storage device of claim 32, wherein said method further comprises analyzing a current state of performance of said computer network system after a predetermined number of throttling steps.

34. The program storage device of claim 33, wherein token issue rates for one or more said client computers are increased by unthrottling their workloads when system resources become available.

35. A computer-implemented system for allocating shared resources in a computer network-based storage system comprising:

means for taking periodic performance samples on a running computer network-based storage system;

means for evaluating an objective function that takes as input said performance samples to quantify how aligned a current state of said computer network-based storage system is with organizational objectives;

means for building and maintaining models of behavior and capabilities of said computer network-based storage system by using said performance samples as input;

means for determining how resources of said computer network-based storage system should be allocated among client computers in said computer network-based storage system by selecting one among many possible allocations based on predictions generated by said models in order to maximize a value of said objective function;

means for calculating a confidence statistic value for a chosen resource allocation based on an accuracy of said models; and

means for enforcing said chosen resource allocation on the running computer network-based storage system when said confidence statistic value is at or above a predetermined threshold value.