System and method for the distribution of a program among cooperating processing elements
AssigneeMANAGEMENT SERVICES GROUP, INC. d/b/a Global Technical Systems
InventorsRobert Stephen Gordy, Terry Spitzer
Life of this patent
Estimating the term…
Abstract
A Veil program analyzes the source code and/or data of an existing sequential target program without user interaction and determines how best to distribute the target program and data among the processing elements of a multi-processing element computing system. The Veil program analyzes source code loops, data sizes and types to prepare a set of distribution attempts or strategies, whereby each strategy is run under a run-time evaluation system and evaluated to determine the optimal decomposition and distribution across the available processing elements.
Claims20
A system that automatically decomposes an existing sequential program into one or more distributed programs that execute on multiple-processing elements of a computing system;
- decomposition is performed without the knowledge of a user, the system comprising:
- a computer system comprising a plurality of processing elements;
- a target program in object code form, the target program written to run on a single processing element computer system;
- a plurality of strategies, the strategies including one or more of parallelization strategies and distribution strategies;
- a Veil program having coded therein a means for:
- (a) encapsulating the target program with a run-time analyzer;
- (b) selecting one of the strategies from the plurality of strategies;
- (c) parallelization and distributing the target program according to the selected one strategy;
- (d) executing the target program by one or more of the processing elements;
- (e) recording execution results of the target program;
- and
- (f) if there are further strategies in the plurality of strategies, changing the selected one strategy to a next strategy from the plurality of strategies and repeat steps a-f;
- (g) comparing the execution results and selecting a best parallelization strategy having a best performance-oriented result;
- and
- (h) parallelizing the target application using the best parallelization strategy with the best performance-oriented result.
A computer implemented method for decomposing and distributing a target program onto a plurality of processing elements without user intervention, the target program being sequential, the method comprising:
- (a) encapsulating the target program with a run-time analyzer;
- (b) selecting one strategy from a plurality of strategies, the strategies being performance oriented parallelization/distribution strategies;
- (c) decomposing and distributing the target program according to the one strategy;
- (d) executing the target program with the run-time analyzer on at least two of the plurality of processing elements;
- (e) recording execution results of the target program;
- and
- (f) if there are further strategies in the plurality of strategies, changing the one strategy to a next strategy from the plurality of strategies and repeating steps a-f.
A system for optimizing the operation of a program in a multiple-processing element computing system, the system comprising:
- (a) a computer system comprising a plurality of processing elements;
- (b) a sequential target program written to run on the computer system;
- (c) a means for encapsulating the target program with a means for analyzing run-time performance of the target program;
- (d) a means for selecting one strategy from a plurality of strategies;
- (e) a means for decomposing and parallelizing the target program according to the one strategy;
- (f) a means for executing the target program with the means for analyzing on one or more of the processing elements;
- (g) a means for recording execution results of the target program;
- and
- (h) if there are further strategies in the plurality of strategies, a means for changing the one strategy to a next strategy from the plurality of strategies and repeat steps c-h;
- (i) a means for comparing the execution results and for selecting a best strategy having a best execution time;
- and
- (j) a means for decomposing and parallelizing the target application using the best strategy.
Dependencies are inferred from the claim text.