Pengembangan Sistem Pengukur Indeks Beban untuk Sistem Penyeimbang Beban pada Komputer Paralel Cluster

Abstract

The performance of a parallel system can be improved in various ways; one of them is load balancing. To perform load balancing task, some load balancing algorithms require the information of load index on each slave computer. This research developed a system to read the load index on each slave computer of a computer cluster that can measure, send, and manage information of load value. The application was implemented using Hyperic SIGAR (System Information Gatherer And Reporter) with CPU, memory, and I/O as the load indices. Analysis of the system performance testing showed that throughput values are proportional to the number of slave computers and inversely proportional to delay. Meanwhile, analysis of load value validity showed that load value is inversely proportional to the delivery delay and the number of slave computer. However, there were several anomalies in the delta analysis due to several factors, such as the unisolated network during the testing period.