A Heterogeneous Processor Set Selection Advistory System for Distributed Data Parallel
碩士 === 國立中央大學 === 資訊及電子工程研究所 === 83 === The elapsed time of executing a data parallel program on a heterogeneous workstations is influenced by data partition and selection. However, data partition and processor's selection is a problem...
| Main Authors: | , |
|---|---|
| Other Authors: | |
| Format: | Others |
| Language: | zh-TW |
| Published: |
1995
|
| Online Access: | http://ndltd.ncl.edu.tw/handle/97137610743855325090 |
| Summary: | 碩士 === 國立中央大學 === 資訊及電子工程研究所 === 83 === The elapsed time of executing a data parallel program on a
heterogeneous workstations is influenced by data partition and
selection. However, data partition and processor's selection is
a problem for the user. .We have developed an method to help
user select processor and redistribute data to reduce the
program's Our method decide the size of granularity by the
number ofhe program's communication. Besides, we redistribute
data to achieve load balance by considering processor's speed
and load.The desire to find the trade-off between the
performance improvement communication overhead due to the
increase of processors. This select processors by estimating
the data parallel program'sur method utilizes information about
machine's speed and machine's design a load monitor to provide
load information. And we assume get program's information about
computation time andime from user. Finally, we show some
experimental results to verify proposed method can reduce the
elapsed time or not.
|
|---|