Replica Location Cost Estimation for Replica Performance in Grid Systems

Main Article Content

P.Sunil Gavaskar
Ch.D.V. Subbaro

Abstract

The objective challenge in High performance computing is fault tolerance and avoidance. In widely used Existing check pointing schemes provides a way of fault detection and recovery. Fault tolerance, communication, efficiency and reliability are important requirements in grid environment. In general replicas are used as proactive (i.e. failure considered before scheduling of a job) and post active handles the job failure after it has occurred. In our proposed model applications job schedule is periodically uses replicas to maximize the success rate of long execution jobs. In order to make the application execution more reliable we use fault tolerant agents replica as centralized and local replicas within agents. Generally static fault tolerance strategies in job scheduling may use additional resources when compare to dynamic strategy. In this paper we plan to introduce the dynamic fault tolerance architecture model for agents oriented replica to overcome the existing fault tolerance uncertainties that occurs due to the parameters like ability of CPU processing, replicas updating cost, replication read cost, replicas search rate. The proposed fault tolerance architecture model is able to setup some parameters of replicas such as centralized and local replicas with varying mean time failures. This model uses number of replicas, replication objects with different states, requirement of task such as replicas availability, processing capability, and mapping of agents replica to tasks. Finally we observed that the replicas utilization and their reliable activation of replication cost are the major activities to provide qualitative fault tolerance model. The proposed model is evaluated using replicas access rate with different mean failure rates.

 

Keywords: Read cost, Search cost, Agent replica, Grid information service, Mean time to failure, Resource fault occurrence history.

Downloads

Download data is not yet available.

Article Details

Section
Articles