A parallel decomposition algorithm for Staircase Linear Programs

Technical Report · Thu Dec 01 04:00:00 EST 1988

OSTI ID:6391734

Entriken, R

As part of an extended research project on the parallel decomposition of linear programs, a parallel algorithm for Staircase Linear Programs was designed and implemented. This class of problems encompasses a large range of planning problems and when decomposed has simple subproblem formulations and communication patterns. This makes its solution a manageable step toward our eventual goal of producing a general code that automatically exploits problem structures of various forms. The results presented here were derived from an implementation for a Sequent Balance 8000 shared-memory multiprocessor. The algorithm itself is message-based but can run on either shared- or distributed-memory parallel computers. A simple diet planning problem is used to demonstrate the principles of the algorithm's development and performance. When applied to this problem, the parallel decomposition algorithm shows promise relative to present serial optimization codes. The nonlinear optimization code MINOS 5.1 is used both as a basis for comparison and as a generic subproblem solver. The greatest room for speedup is in exploiting problem structures. The results show that decomposition can improve efficiency even with a single processor. Examples are given where multiple processors lead to still greater efficiency. 9 refs., 13 figs., 4 tabs.

• Research Organization: Oak Ridge National Lab., TN (USA)
• DOE Contract Number: AC05-84OR21400
• OSTI ID: 6391734
• Report Number(s): ORNL/TM-11011; ON: DE89007883
• Country of Publication: United States
• Language: English

Related Subjects

99 GENERAL AND MISCELLANEOUS
990210* -- Supercomputers-- (1987-1989)
ALGORITHMS
ARRAY PROCESSORS
EFFICIENCY
LINEAR PROGRAMMING
MATHEMATICAL LOGIC
PARALLEL PROCESSING
PERFORMANCE
PROGRAMMING