Resource discovery in computational grids: quantitative comparison and analysis of MDS and DHT

Abstract

Grid Information Services provided by most of the grid middlewares aim at discovering resources in wake of large number of Grid nodes. Many Grid Information Services (GIS) have been deployed in different Grid middlewares. While deploying a Grid, it is a challenging task for the Grid administrator to opt for the GIS which performs better. Till date not many results exist to support such decision making. This work aims to provide the behavior and performance of two most widely deployed GIS viz. Monitoring and Discovery Services (MDS) and Distributed Hash Tables (DHT). A quantitative comparative analysis of the two schemes has been performed. Experiments have been designed and performed over GridSim, to understand the behavior of the two services in presence of large number of users and resources. Four experiments are conducted, scalability of the two schemes has been studied, and the parameters that form bottlenecks in the behavior of these two schemes have been identified. Upper bounds have been established over the parameters identified, e.g. number of users an index server can support, number of information providers, response time and throughput which affect the performance. Finally the results obtained establish that DHT index server scales better and provide higher throughput as compared to MDS index server. The results provide an aid in deploying of the information services and help in future development work.