Factors affecting the design of load balancing algorithms in distributed systems

In a distributed environment made up of several computing units, it is necessary to exploit the system's greater calculation resources more efficiently by distributing the workload over the various machines in the system. The aim of distributed allocation can therefore be to balance the load generated in the single machines so as to improve the response time and throughput of the system as a whole. This paper deals with the problem of allocating processes with the aim of achieving load balancing in a heterogeneous distributed system. Some strategies are proposed based on adaptive heuristic approaches which can be followed in real systems. The aim is to point out which factors mainly influence a good workload distribution and which parameters can be tuned to achieve the best allocation. In order to measure the host's workload here the concept of relaxation is introduced, and some algorithms which take this parameter into account are discussed. In addition, the contribution of the knowledge of the process nature to the allocation strategy has been investigated too. The solutions proposed have all been implemented and evaluated on a real heterogeneous distributed system in order to assess how each parameter, in combination with the others, contributes to determining the best allocation decision. |