Cluster-based load partitioning and allocation in distributed computing systems

Abstract

Most of the task allocation models and algorithms in distributed computing system (DCS) require a priori knowledge of the execution time of task on the processing nodes of the DCS. Since the task assignment is not known in advance, the execution time is difficult to estimate. We propose a cluster-based load partitioning and allocation, in a distributed computing system, that eliminates the need to know the execution time of the task a priori. It considers the allocation by clustering both the task and the nodes of the DCS. The clustering of modules of a task and the clustering of processing nodes is done using a fuzzy function. The fuzzy functions are based on the communication requirement of the modules for module clustering and on the interprocessor distance for processor clustering. The algorithm uses dynamic invocation of clustering and assignment routines. A simple example is illustrated to show the clustering and allocation process. Experimental results are presented to support the allocation.