Stochastic analysis of gang scheduling in parallel and distributed systems

被引：0

作者：

Squillante, Mark S. ^{[1
]}

Wang, Fang ^{[1
]}

Papaefthymiou, Marios ^{[1
]}

机构：

[1] IBM Thomas J. Watson Research Cent, Yorktown Heights, United States

来源：

| 1996年 / Elsevier Science B.V., Amsterdam, Netherlands卷 / 27-28期

关键词：

Mathematical models - Numerical analysis - Queueing theory - Random processes - Resource allocation - Response time (computer systems) - Time sharing systems;

D O I：

10.1016/0166-5316(96)00042-9

中图分类号：

学科分类号：

摘要：

Gang scheduling is an approach for resource allocation in parallel and distributed systems that combines time-sharing with space-sharing to ensure a short response time for interactive tasks and high overall system throughput. In this paper, we present queueing theoretic models for a particular gang scheduling system under a workload representative of large-scale engineering and scientific computing environments. We derive a detailed mathematical analysis of these models, from which we obtain closed-form expressions for different performance measures of interest. Our model and analysis is then used to analyze several fundamental performance tradeoffs associated with gang scheduling.

引用