P. Huttunen, J. Ikonen, and J. Porras (Finland)
Parallel simulation, scheduling, data-parallelism, performance optimization, GSM network.
This paper describes a research project where a sequential GSM network simulator was parallelized in a data-parallel fashion. Initially, three scheduling algorithms were implemented to handle the distribution of workunits among processors. These three algorithms were later redesigned and implemented by providing the algorithms with application specific information. By application-specific information we consider information about the workunits such as computational requirements of the workunits. The application-specific information is either given by the user or extracted from the application during run-time. The additional information provided to the scheduling algorithms proved to have a significant impact on their capability to produce more optimal work balance among the processors. The results show that workload variation was reduced by up to 90%.
Important Links:
Go Back
