High Discovery Proportion and Fault Tolerance of Grid Information Service

Abstract

Resource Information Service has been one of the fundamental issues in the Grid computing environment. A resource information mechanism should efficiently manipulate the registering, discovering (querying), updating, etc. of grid resources. Various approaches about resource discovery have been suggested, which are either centralized or hierarchical. They operate inefficiently as the scale of grid systems rapidly increases. In this paper, we propose a distributed GIS system (QFIS) that is based on our Legion model and deploys a number of registry servers at the service region. With a small quorum size √N and the symmetric property, the Leg-Ring scheme can be extended to a fault tolerant and load balanced GIS algorithm. Due to the characteristics of our Legion structure, the user could obtain 100 percent of resource information satisfying the query constraints without the need to search all GIS servers, even if some of them crashed. In addition, we have completed the optimality proof of Legion, which indicates the minimal quorum size in our scheme. Moreover, the cost analysis evaluates the performances of QFIS model and typical hierarchical (HRIS) model and shows that QFIS performance is between the performances of H = 3 (tree height) and H = 4 of HRIS. Finally, the simulation analysis also illustrates that our proposed QFIS can work better than HRIS and possess fault tolerant capability.