S.F. EL-Zoghdy, H. Kameda, and J. Li (Japan)
Distributed computer systems, Static loadbalancing, Dynamic load balancing, Individually optimalpolicies, Performance evaluation.
This study focuses on performance comparison between static and dynamic individually load balancing policies in a distributed computer system where truly optimal so lutions of both dynamic and static policies have been characterized. In this paper, we consider a processor sharing service system (a mainframe node) which serves all its users (jobs) simultaneously with service rate to in dividual jobs decreasing as the load increases. Each ar riving job may observe the current load and then choose whether to join the shared system. The alternative is a personal computer node that offers a fixed expected ser vice time. The processes wish to minimize their individ ual service times. However, the optimal choice for each job depends on the decisions of subsequent ones, through their effect on the future load in the shared server. The overheads and the delays for both job transfer and system state-information exchange are assumed to be negligible. The results show that, in the model examined, the dy namic individually optimal load balancing policy outper forms the static one in the overall mean response time, at most about 48 percent in the range of parameter values examined. The difference is of a certain magnitude for the cases where the arrival rate to the system is close to the processing rate of the shared system node for rather large values of both. Also, for these cases, the difference between the ratio that an arriving job to the system goes to the shared node under the static and dynamic load bal ancing policies is of a certain magnitude.
Important Links:
Go Back
