Allocation of resources among computer partitions using plural utilization prediction engines

Patent metadata

Field	Value
Publication number	US-9195508-B1
Application number	US-74573707-A
Country	US
Kind code	B1
Filing date	May 8, 2007
Priority date	May 8, 2007
Publication date	Nov 24, 2015
Grant date	Nov 24, 2015

Abstract

A multi-partition computer system includes a utilization monitor for monitoring resource utilization, plural prediction engines for predicting utilization during a next allocation period, a prediction rater for rating said prediction engines based on the accuracy of their predictions, and an allocation implementer for implementing an allocation determined as a function of a prediction by a highest-rated of said prediction engines as determined by the prediction rater.

First claim

What is claimed is:

1. A computer system comprising:
plural computer partitions to run respective workloads, wherein at least one of the plural computer partitions comprises at least one processor;
a utilization monitor to monitor computer resource utilization by said workloads to yield resource utilization data;
plural prediction engines to generate competing resource-demand predictions for a next allocation period, each of said prediction engines making a resource-demand prediction for resources to be required by each of said workloads for said next allocation period;
a prediction-engine rater to calculate prediction penalties for the plural prediction engines using a penalty function, and to determine a best-performing prediction engine and one or more non-best-performing prediction engines for said next allocation period based on the prediction penalties, wherein a shape of the penalty function is specified by a user-settable shape parameter; and
an allocation implementer to implement an allocation for said next allocation period, said allocation being based on the prediction made by said best-performing prediction engine and not on a prediction made by a non-best-performing prediction engine.

2. A computer system as recited in claim 1 wherein the shape of the penalty function indicates prediction penalties associated with discrepancies between respective resource-demand predictions and actual utilization values determined by said utilization monitor for said next allocation period.

3. A computer system as recited in claim 1 further comprising a prediction selector for selecting the prediction by said best-performing prediction engine for implementation by said allocation implementer for said next allocation period.

4. A computer system as recited in claim 1 wherein at least one of said prediction engines generates a non-historical prediction that is not based on utilization for an allocation period or periods preceding a current allocation period immediately preceding said next allocation period.

5. A computer system as recited in claim 1 wherein at least one of said prediction engines predicts based on extrapolating trends in utilization.

6. A computer system as recited in claim 1 further comprising a prediction selector for selecting the prediction of said best-performing prediction engine, said prediction-engine rater rating the output of said prediction selector along with the outputs of said prediction engines.

7. A computer system as recited in claim 1 wherein said allocation implementer allocates at least one of processors, storage media, and communications devices among partitions.

8. A computer system as recited in claim 1 wherein the shape of the penalty function indicates prediction penalties associated with deviations from optimum resource allocations as determined from said resource-utilization data.

9. A computer system as recited in claim 1 wherein the shape of the penalty function indicates a prediction penalties associated with combinations of deviations of actual utilization from predicted utilization and deviations from optimum resource allocations as determined from resource-allocation data.

10. A computer system as recited in claim 1 wherein a range and coefficients associated with the penalty function are adjustable through user-settable parameters.

11. A computer workload management method comprising:
plural prediction engines generating competing respective computer resource-demand predictions for an allocation period, each of said predictions predicting a resource demand for each of plural workloads;
calculating prediction penalties for the plural prediction engines using a penalty function, wherein a shape of the penalty function is specified by a user-settable shape parameter;
selecting, using at least one hardware processor, one of said predictions based on the prediction penalties so that the remaining of said predictions are non-selected predictions for said allocation period; and
implementing an allocation of computer resources to workloads for said allocation period, said allocation being a function of the selected prediction and not as a function of one of said non-selected predictions.

12. A method as recited in claim 11 further comprising:
evaluating an accuracy of each of said predictions based on actual utilization during said allocation period, wherein the shape of the penalty function indicates prediction penalties associated with accuracies of predictions.

13. A method as recited in claim 12 wherein said selecting comprises selecting a prediction from a prediction engine based on the prediction penalties over a plurality of allocation periods.

14. A method as recited in claim 13 wherein the shape of the penalty function indicates a prediction penalties associated with combinations of deviations of actual utilization from predicted utilization and deviations from optimum resource allocations as determined from resource-allocation data.

15. A method as recited in claim 11 wherein utilization data used by a first of said prediction engines in generating a prediction for said allocation period is accumulated over a longer duration than is utilization data used by a second of said prediction engines in generating its predictions for said allocation period.

16. A method as recited in claim 15 wherein a third of said prediction engines considers data other than time and utilization in generating its predictions.

17. A method as recited in claim 11 wherein said implementing involves reallocating at least one computer resource selected from a set including processors, memory, and communications devices from one workload to another workload.

18. A computer product comprising non-transitory computer-readable storage media encoded with code defining:
plural computer-executable prediction engines to generate competing respective computer-resource-demand predictions for a single allocation period, all of said resource-demand predictions specifying amounts of resource predicted to be demanded by a first set of plural workloads for said allocation period;
a prediction-engine rater executable by at least one hardware processor to calculate prediction penalties for the plural computer-executable prediction engines using a penalty function, wherein a shape of the penalty function is specified by a user-settable shape parameter;
a computer-executable prediction selector for selecting one of said predictions based on the prediction penalties so that the one or more others of said predictions are non-selected predictions; and
a computer-executable implementer for directing implementation of an allocation of computer resources to workloads for said allocation period, said allocation being a function of the selected prediction and not of said non-selected predictions.

19. A computer product as recited in claim 18 , said code further defining:
a computer-executable prediction evaluator for evaluating an accuracy for each of said predictions.

20. A computer product as recited in claim 18 , wherein the shape of the penalty function indicates a prediction penalties associated with combinations of deviations of actual utilization from predicted utilization and deviations from optimum resource allocations as determined from resource-allocation data.

21. A computer product as recited in claim 18 wherein the user-settable shape parameter comprises at least a linear shape and a quadratic shape.

22. A computer product as recited in claim 18 w

Assignees

• Blanding William H
• Singhal Sharad
• Hewlett Packard Development Co

Inventors

• Blanding William H
• Singhal Sharad

Classifications

• G06F9/505 Primary

  considering the load · CPC title

• G06F9/5027 Primary

  the resource being a machine, e.g. CPUs, Servers, Terminals · CPC title

• G06F2209/5019

  Workload prediction · CPC title

• Y02D10/00

  Energy efficient computing, e.g. low power processors, power management or thermal management · CPC title