Runtime optimization of an application executing on a parallel computer

• US 8,898,678 B2
• Filed: 10/30/2012
• Issued: 11/25/2014
• Est. Priority Date: 04/14/2010
• Status: Expired due to Fees

• Current Assignee: International Business Machines Corporation
• Original Assignee: International Business Machines Corporation
• Inventors: Faraj, Daniel A., Smith, Brian E.
• Primary Examiner(s): Lee, Adam
• Application Number:
  US13/663,545
• Publication Number:
  US 20130067487A1
• Time in Patent Office: 756 Days
• Field of Search: None
• US Class Current: 718/106
• CPC Class Codes: G06F 11/3404 for parallel or distributed programming
• CPC Class Codes: G06F 9/52 Program synchronisation; Mutual exclusion, e.g. by means of semaphores
• CPC Class Codes: G06F 9/546 Message passing systems or structures, e.g. queues

Abstract

Identifying a collective operation within an application executing on a parallel computer; identifying a call site of the collective operation; determining whether the collective operation is root-based; if the collective operation is not root-based: establishing a tuning session and executing the collective operation in the tuning session; if the collective operation is root-based, determining whether all compute nodes executing the application identified the collective operation at the same call site; if all compute nodes identified the collective operation at the same call site, establishing a tuning session and executing the collective operation in the tuning session; and if all compute nodes executing the application did not identify the collective operation at the same call site, executing the collective operation without establishing a tuning session.

Claims

1. An apparatus for runtime optimization of an application executing on a parallel computer, the parallel computer having a plurality of compute nodes organized into a communicator, the apparatus comprising a computer processor and a computer memory operatively coupled to the computer processor, the computer memory having disposed within it computer program instructions that when executed by the computer processor cause the apparatus to carry out the steps of:

determining, by each compute node, whether a collective operation is root-based;

if the collective operation is not root-based, establishing a tuning session administered by a self tuning module for the collective operation in dependence upon an identifier of a call site of the collective operation and executing the collective operation in the tuning session;

if the collective operation is root-based, determining, through use of a single other collective operation, whether all compute nodes executing the application identified the collective operation at the same call site;

if all compute nodes executing the application identified the collective operation at the same call site, establishing a tuning session administered by the self tuning module for the collective operation in dependence upon the identifier of the call site of the collective operation and executing the collective operation in the tuning session; and

if all compute nodes executing the application did not identify the collective operation at the same call site, executing the collective operation without establishing a tuning session.

2. The apparatus of claim 1 wherein a root-based collective operation comprises one of:

a broadcast operation, a scatter operation, a gather operation, or a reduce operation.

3. The apparatus of claim 1 wherein determining whether all compute nodes executing the application identified the collective operation at the same call site further comprising performing on all the compute nodes of the communicator an ‘

allreduce’

collective operation to identify the minimum and maximum values of all of the identified call sites.

4. The apparatus of claim 1 further comprising computer program instructions that when executed by the computer processor cause the apparatus to carry out the steps of:

selecting, for a particular collective operation of the application in dependence upon one or more tuning sessions for the particular collective operation, one or more algorithms to carry out the particular collective operation, the one or more algorithms representing an optimized set of algorithms to carry out the particular collective operation;

recording the one or more selected algorithms; and

during a subsequent execution of the application and without performing another tuning session, carrying out the particular collective operation of the application with the recorded selected algorithms.

5. The apparatus of claim 4 wherein recording the one or more selected algorithms from the tuning session further comprises recording, in association with the one or more selected algorithms, an identifier of the call site for the particular collective operation, a message size, and a communicator identifier.

6. The apparatus of claim 4 wherein:

recording the one or more selected algorithms from the tuning session further comprises identifying any of the tuned collective operations that are non-critical collective operations; and

carrying out the particular collective operation of the application with the recorded selected algorithms further comprises carrying out the non-critical collective operations with standard messaging module algorithms.

7. A computer program product for runtime optimization of an application executing on a parallel computer, the parallel computer having a plurality of compute nodes organized into a communicator, the computer program product disposed in a computer readable hardware storage medium, the computer program product comprising computer program instructions that when executed by a processor cause a computer to carry out the steps of:

determining, by each compute node, whether a collective operation is root-based;

if the collective operation is not root-based, establishing a tuning session administered by a self tuning module for the collective operation in dependence upon an identifier of a call site of the collective operation and executing the collective operation in the tuning session;

if the collective operation is root-based, determining, through use of a single other collective operation, whether all compute nodes executing the application identified the collective operation at the same call site;

if all compute nodes executing the application identified the collective operation at the same call site, establishing a tuning session administered by the self tuning module for the collective operation in dependence upon the identifier of the call site of the collective operation and executing the collective operation in the tuning session; and

if all compute nodes executing the application did not identify the collective operation at the same call site, executing the collective operation without establishing a tuning session.

8. The computer program product of claim 7 wherein a root-based collective operation comprises one of:

a broadcast operation, a scatter operation, a gather operation, or a reduce operation.

9. The computer program product of claim 7 wherein determining whether all compute nodes executing the application identified the collective operation at the same call site further comprising performing on all the compute nodes of the communicator an ‘

allreduce’

collective operation to identify the minimum and maximum values of all of the identified call sites.

10. The computer program product of claim 7 further comprising computer program instructions that when executed by a processor cause a computer to carry out the steps of:

selecting, for a particular collective operation of the application in dependence upon one or more tuning sessions for the particular collective operation, one or more algorithms to carry out the particular collective operation, the one or more algorithms representing an optimized set of algorithms to carry out the particular collective operation;

recording the one or more selected algorithms; and

during a subsequent execution of the application and without performing another tuning session, carrying out the particular collective operation of the application with the recorded selected algorithms.

11. The computer program product of claim 10 wherein recording the one or more selected algorithms from the tuning session further comprises recording, in association with the one or more selected algorithms, an identifier of the call site for the particular collective operation, a message size, and a communicator identifier.

12. The computer program product of claim 10 wherein:

recording the one or more selected algorithms from the tuning session further comprises identifying any of the tuned collective operations that are non-critical collective operations; and

carrying out the particular collective operation of the application with the recorded selected algorithms further comprises carrying out the non-critical collective operations with standard messaging module algorithms.