Method and apparatus for reducing execution time for parallel support vector machine computations

• US 20070179927A1
• Filed: 01/27/2006
• Published: 08/02/2007
• Est. Priority Date: 01/27/2006
• Status: Active Grant

• Current Assignee: Oracle America, Inc. (Oracle Corporation)
• Original Assignee: Oracle America, Inc. (Oracle Corporation)
• Inventors: Vaidyanathan, Kalyanaraman, Gross, Kenny
• Granted Patent: US 7,865,898 B2
• Time in Patent Office: Days
• US Class Current: 1/1
• CPC Class Codes: G06F 18/2411 based on the proximity to a decision surface, e.g. support vector machines
• CPC Class Codes: G06V 10/94 Hardware or software architectures specially adapted for image or video understanding

Abstract

A system that reduces execution time of a parallel SVM application. During operation, the system partitions an input data set into chunks of data. Next, the system distributes the partitioned chunks of data across a plurality of available computing nodes and executes the parallel SVM application on the chunks of data in parallel across the plurality of available computing nodes. The system then determines if a first timeout period has been exceeded before all of the plurality of available computing nodes have finished processing their respective chunks of data. If so, the system (1) repartitions the input data set into different chunks of data; (2) redistributes the repartitioned chunks of data across some or all of the plurality of available computing nodes; and (3) executes the parallel SVM application on the repartitioned chunks of data in parallel across some or all of the available computing nodes.

Claims

1. A method for reducing execution time of a parallel support vector machine (SVM) application, comprising:

partitioning an input data set into chunks of data;

distributing the partitioned chunks of data across a plurality of available computing nodes;

executing the parallel SVM application on the chunks of data in parallel across the plurality of available computing nodes;

determining if a first timeout period has been exceeded before all of the plurality of available computing nodes have finished processing their respective chunks of data; and

if so,

repartitioning the input data set into different chunks of data;

redistributing the repartitioned chunks of data across some or all of the plurality of available computing nodes; and

executing the parallel SVM application on the repartitioned chunks of data in parallel across some or all of the available computing nodes.

2. The method of claim 1, wherein if the first timeout period has been exceeded before all of the plurality of available computing nodes have finished processing their respective chunks of data, prior to repartitioning the input data set into different chunks of data, the method further comprises:

cancelling the execution of the parallel SVM application; and

deleting results of the parallel SVM application.

3. The method of claim 1, wherein determining if the first timeout period has been exceeded before all available computing nodes have finished processing their respective chunks of data involves:

determining if a portion of the plurality of available computing nodes have completed processing associated chunks of data;

if not, waiting until a portion of the plurality of available computing nodes have completed processing their respective chunks of data; and

otherwise,

computing the mean of the completion times for the portion of the plurality of available computing nodes that have completed processing their respective chunks of data; and

setting the first timeout period to a constant factor times the mean of the completion times minus the current elapsed time.

4. The method of claim 3, wherein the constant factor is three.

5. The method of claim 1, wherein if the first timeout period has not been exceeded before all available computing nodes have finished processing their respective chunks of data, the method further comprises:

combining the results from each processed chunk of data;

executing the parallel SVM application to process the combined results;

computing the mean of the completion time for all chunks of data;

setting a second timeout period to a constant factor times the mean of the completion times for all chunks of data;

determining if the second timeout period has been exceeded before the parallel SVM application has completed processing the combined results;

if so,

cancelling the execution of the parallel SVM application on the combined results;

deleting results of the parallel SVM application;

repartitioning the input data set into different chunks of data;

redistributing the repartitioned chunks of data across the plurality of available computing nodes; and

executing the parallel SVM application on the repartitioned chunks of data in parallel across some or all of the available computing nodes; and

otherwise, outputting the results.

6. The method of claim 1, wherein repartitioning the input data set into different chunks of data involves:

removing one computing node from the plurality of available computing nodes; and

repartitioning the input data set into chunks of data for the remaining computing nodes in the plurality of available computing nodes.

7. The method of claim 1, wherein if only a single computing node is available, the method further comprises:

executing the parallel processing application on the single computing node to process the input data set; and

outputting the result.

8. The method of claim 1, wherein partitioning the input data set into chunks of data involves partitioning the input data set so that the number of chunks of data equals the number of available computing nodes.

9. A computer-readable storage medium storing instructions that when executed by a computer cause the computer to perform a method for reducing execution time of a parallel support vector machine (SVM) application, the method comprising:

partitioning an input data set into chunks of data;

distributing the partitioned chunks of data across a plurality of available computing nodes;

executing the parallel SVM application on the chunks of data in parallel across the plurality of available computing nodes;

determining if a first timeout period has been exceeded before all of the plurality of available computing nodes have finished processing their respective chunks of data; and

if so,

repartitioning the input data set into different chunks of data;

redistributing the repartitioned chunks of data across some or all of the plurality of available computing nodes; and

executing the parallel SVM application on the repartitioned chunks of data in parallel across some or all of the available computing nodes.

10. The computer-readable storage medium of claim 9, wherein if the first timeout period has been exceeded before all of the plurality of available computing nodes have finished processing their respective chunks of data, prior to repartitioning the input data set into different chunks of data, the method further comprising:

cancelling the execution of the parallel SVM application; and

deleting results of the parallel SVM application.

11. The computer-readable storage medium of claim 9, wherein determining if the first timeout period has been exceeded before all available computing nodes have finished processing their respective chunks of data involves:

determining if a portion of the plurality of available computing nodes have completed processing associated chunks of data;

if not, waiting until a portion of the plurality of available computing nodes have completed processing their respective chunks of data; and

otherwise,

computing the mean of the completion times for the portion of the plurality of available computing nodes that have completed processing their respective chunks of data; and

setting the first timeout period to a constant factor times the mean of the completion times minus the current elapsed time.

12. The computer-readable storage medium of claim 11, wherein the constant factor is three.

13. The computer-readable storage medium of claim 9, wherein if the first timeout period has not been exceeded before all available computing nodes have finished processing their respective chunks of data, the method further comprising:

combining the results from each processed chunk of data;

executing the parallel SVM application to process the combined results;

computing the mean of the completion time for all chunks of data;

setting a second timeout period to a constant factor times the mean of the completion times for all chunks of data;

determining if the second timeout period has been exceeded before the parallel SVM application has completed processing the combined results;

if so,

cancelling the execution of the parallel SVM application on the combined results;

deleting results of the parallel SVM application;

repartitioning the input data set into different chunks of data;

redistributing the repartitioned chunks of data across the plurality of available computing nodes; and

executing the parallel SVM application on the repartitioned chunks of data in parallel across some or all of the available computing nodes; and

otherwise, outputting the results.

14. The computer-readable storage medium of claim 9, wherein repartitioning the input data set into different chunks of data involves:

removing one computing node from the plurality of available computing nodes; and

repartitioning the input data set into chunks of data for the remaining computing nodes in the plurality of available computing nodes.

15. The computer-readable storage medium of claim 9, wherein if only a single computing node is available, the method further comprising:

executing the parallel processing application on the single computing node to process the input data set; and

outputting the result.

16. The computer-readable storage medium of claim 9, wherein partitioning the input data set into chunks of data involves partitioning the input data set so that the number of chunks of data equals the number of available computing nodes.

17. An apparatus that reduces execution time of a parallel support vector machine (SVM) application, comprising:

a partitioning mechanism configured to partition an input data set into chunks of data;

a distributing mechanism configured to distribute the partitioned chunks of data across a plurality of available computing nodes;

an execution mechanism configured to execute the parallel SVM application on the chunks of data in parallel across the plurality of available computing nodes;

a timeout mechanism configured to;

determine if a first timeout period has been exceeded before all of the plurality of available computing nodes have finished processing their respective chunks of data; and

if so, to

repartition the input data set into different chunks of data;

redistribute the repartitioned chunks of data across some or all of the plurality of available computing nodes; and

to

execute the parallel SVM application on the repartitioned chunks of data in parallel across some or all of the available computing nodes.

18. The apparatus of claim 17, wherein if the first timeout period has been exceeded before all of the plurality of available computing nodes have finished processing their respective chunks of data, prior to repartitioning the input data set into different chunks of data, the timeout mechanism is configured to:

cancel the execution of the parallel SVM application; and

to

delete results of the parallel SVM application.

19. The apparatus of claim 17, wherein while determining if the first timeout period has been exceeded before all available computing nodes have finished processing their respective chunks of data, the timeout mechanism is configured to:

determine if a portion of the plurality of available computing nodes have completed processing associated chunks of data;

if not, to wait until a portion of the plurality of available computing nodes have completed processing their respective chunks of data; and

otherwise, to

compute the mean of the completion times for the portion of the plurality of available computing nodes that have completed processing their respective chunks of data; and

to

set the first timeout period to a constant factor times the mean of the completion times minus the current elapsed time.

20. The apparatus of claim 17, wherein if the first timeout period has not been exceeded before all available computing nodes have finished processing their respective chunks of data,

the execution mechanism is configured to:

combine the results from each processed chunk of data;

execute the parallel SVM application to process the combined results;

compute the mean of the completion time for all chunks of data; and

to

set a second timeout period to a constant factor times the mean of the completion times for all chunks of data; and

the timeout mechanism is configured to;

determine if the second timeout period has been exceeded before the parallel SVM application has completed processing the combined results;

if so, to

cancel the execution of the parallel SVM application on the combined results;

delete results of the parallel SVM application;

repartition the input data set into different chunks of data;

redistribute the repartitioned chunks of data across the plurality of available computing nodes; and

to

execute the parallel SVM application on the repartitioned chunks of data in parallel across some or all of the available computing nodes; and

otherwise, to output the results.