History-based adaptive load sharing heuristics

Author

Cho, Seung H. ; Choi, Seung R.

Author_Institution

Dept. of Comput. Eng., Kangnam Univ., South Korea

fYear

1994

fDate

22-26 Aug 1994

Firstpage

161

Abstract

The conventional load representation method expressing a load as queue length has a weakness in that it does not reflect execution characteristics of processes. To solve the inaccuracy of this conventional method of load representation, we propose the weighted load representation method which assigns graded weights to processes based on the execution characteristics of processes. The proposed method uses history information obtained during process scheduling. In order to derive the upper bound and the lower bound of the load in a node, the finite population queuing model is used. We devise heuristics for adaptive load sharing using the proposed representation method and derived upper and lower bounds. Simulation results show that GRRs of weight functions log i and i are improved 2.37-3.14 times compared with no load sharing and 14-24% as many as GRRs of the conventional load representation method

Keywords

performance evaluation; processor scheduling; queueing theory; resource allocation; adaptive load sharing heuristics; execution characteristics; finite population queuing model; graded weights; history information; history-based adaptive load sharing; load representation; load representation method; lower bound; process scheduling; queue length; simulation results; upper bound; weight functions; weighted load representation method; Computational modeling; Delay; Distributed computing; History; Load management; Military computing; Pressing; Throughput; Upper bound; Workstations;

fLanguage

English

Publisher

ieee

Conference_Titel

TENCON '94. IEEE Region 10's Ninth Annual International Conference. Theme: Frontiers of Computer Technology. Proceedings of 1994

Print_ISBN

0-7803-1862-5

Type

conf

DOI

10.1109/TENCON.1994.369316

Filename

369316