Maintaining Replica Consistency Over Large-Scale Data Grid Using Update Propagation Technique

A Data Grid is an organized collection of nodes in a wide area network which contributes to various computation, storage data, and application. In Data Grid high numbers of users are distributed in a wide area environment which is dynamic and heterogeneous. Data management is one of the current issues where data transparency, consistency, fault-tolerance, automatic management and the performance are the user parameters in grid environment. Data management techniques must scale up while addressing autonomy, dynamicity and heterogeneity of the data resource. Data replication is a well known technique used to reduce accesses latency, improve availability and performance in a distributed computing environment. Replication introduces the problem of maintaining consistency among the replicas when files are allowed to be updated. The update information should be propagated to all replicas to guarantee correct read of the remote replicas. An asynchronous replication is a commonly agreed solution for the problem in consistency of replicas. A few studies have been done to maintain replica consistency in Data Grid. However, the introduced techniques are neither efficient nor scalable. They cannot be used in real Data Grid since the issues of large number of replica sites, large scale distribution, load balancing and site autonomy where the capability of grid site to join and leave the grid community at any time have not been addressed. This thesis proposes a new asynchronous replication protocol called Update Propagation Grid (UPG) to maintain replica consistency over a large scale data grid. In UPG the updates reach all on-line secondary replicas using a propagation technique based on nodes organized into a logical structure network in the form of two-dimensional grid structure. The proposed update propagation technique is a hybrid push-pull and dynamic technique that addresses the issues of site autonomy, efficiency, scalability, load balancing and fairness. A two performance analysis studies have been conducted to study the performance of the proposed technique in comparison with other techniques. First study involves mathematical and simulation analysis. Second study is based on Queuing Network Model. The result of the performance analysis shows that the proposed technique scales well with high number of replica sites and with high request loads. The result also shows the reduction on the average update reach time by 5% to 97%. Moreover the result shows that the proposed technique is capable of reaching load balancing while providing update propagation fairness

Preview PDF FSKTM_2009_7a.pdf Download (683kB)

Actions (login required)

View Item