Abstracts

A Detailed Comparison of Traditional and Neural Network Based Approaches for Load Metrics on Heterogeneous Platforms

Adaptive and dynamic load balancing are challenging problems in the field of parallel computing. Therefore, a load management facility is desirable which gives up to date load informations which are also comparable for heterogeneous machines. In this paper we discuss several approaches for load metrics in heterogeneous systems. We compare simple load metrics with neural networks which have been trained to predict the expected delay of an application from the sampled load informations. The results show that the proposed load metric performs well in heterogeneous environments. Further, neural networks can improve the performance of load balancing facilities.

[back] [top]
Mail to webmaster