Performance bounds on multiprocessor schedules

Conference · Sun Jan 01 04:00:00 EST 1984

OSTI ID:5082359

Kain, R Y; Raie, A A

Optimizing general uniprocessor task scheduling is an np-hard problem. For multiprocessor systems, list scheduling is not optimum but is easily implemented. Bounds on the total finishing time of list scheduling algorithms are known. This paper proves bounds on the performance of a multiprocessor list scheduling algorithm compared to the optimum schedule for the particular problem instance. It considers situations of which the system performance is measured by the weighted mean task finish time or measured by payoffs that exponentially decrease as a function of task completion times. For each measure, bounds are determined for two cases: (1) all processors can initiate processing at the same time, and (2) each processor has a given start time. All four problems utilize similar proofs. The proofs and results apply when task execution times are known quantities or when they are random variables with known probability distributions; the latter case is not discussed in this paper. 5 references.

• OSTI ID: 5082359
• Country of Publication: United States
• Language: English

Related Subjects

99 GENERAL AND MISCELLANEOUS
990200* -- Mathematics & Computers
ARRAY PROCESSORS
DATA PROCESSING
PERFORMANCE
PROCESSING
TASK SCHEDULING