Speedup properties of phases in the execution profile of distributed parallel programs
- 1 August 1992
- report
- Published by Office of Scientific and Technical Information (OSTI)
Abstract
The execution profile of a distributed-memory parallel program specifies the number of busy processors as a function of time. Periods of homogeneous processor utilization are manifested in many execution profiles. These periods can usually be correlated with the algorithms implemented in the underlying parallel code. Three families of methods for smoothing execution profile data are presented. These approaches simplify the problem of detecting end points of periods of homogeneous utilization. These periods, called phases, are then examined in isolation, and their speedup characteristics are explored. A specific workload executed on an Intel iPSC/860 is used for validation of the techniques described.Keywords
This publication has 0 references indexed in Scilit: