A dynamic load distribution strategy for systems under high task variation and heavy traffic

Fu, B and Tari, Z 2003, 'A dynamic load distribution strategy for systems under high task variation and heavy traffic', in B. Panda (ed.) Proceedings of the 2003 ACM Symposium on Applied Computing, Melbourne, USA, 2003, pp. 1031-1037.


Document type: Conference Paper
Collection: Conference Papers

Title A dynamic load distribution strategy for systems under high task variation and heavy traffic
Author(s) Fu, B
Tari, Z
Year 2003
Conference name ACM Symposium on Applied Computing
Conference location Melbourne, USA
Conference dates 2003
Proceedings title Proceedings of the 2003 ACM Symposium on Applied Computing
Editor(s) B. Panda
Publisher ACM Press
Place of publication New York, USA
Start page 1031
End page 1037
Abstract Several approaches have been proposed to deal with the issue of load distribution, however they all have similar limitations, such as: (i) tasks are executed in an arbitrary order (which may cause large tasks to be delayed), (ii) the task dispatcher does not take into consideration the server processing capacity (which may cause a large task to be assigned to a server with low processing power) or (iii) they do not consider task deadlines (which if not met, may cause task starvation). This paper proposes an extension of LFF (Least Flow-time First) task assignment policy [9], called LFF-PRIORITY, to deal with these limitations. LFF-PRIORITY dynamically computes two priorities, namely task size and task size priorities, and put them in a priority based multi-section queue. The testing results clearly show that LFF-PRIORITY out performs existing load distribution strategies (that are based on heavy tailed distribution). The testing results also show that more than 80% of tasks meet their task deadline under LFF-PRIORITY.
Subjects Distributed and Grid Systems
Keyword(s) design
measurement
performance
theory
Copyright notice © 2003 ACM
ISBN 1581136242
Versions
Version Filter Type
Access Statistics: 223 Abstract Views  -  Detailed Statistics
Created: Thu, 01 Apr 2010, 11:34:28 EST by Catalyst Administrator
© 2014 RMIT Research Repository • Powered by Fez SoftwareContact us