In heterogeneous data processing a set of processors, with possibly different storage and processing capacities are networked to satisfy the user requirements. The network is scalable and distributed. When a user request for an item of data is received, it needs to be serviced with a low access latency and high availability. Data replication is one of the techniques used by the network to to achieve high availability, reliability and fault tolerance. Replication involves storing multiple copies of the data being stored at different locations. How many copies to keep is the replication factor, and where to keep is the replica placement issue, and when to delete unnecessary copies is replica eviction. This paper deals with the issue of determining the replication factor based on the popularity (or request rate) of the particular data. Replicas are placed based on the user request patterns. We have proposed a dynamic adaptive data replication scheme to place the data items in different locations based on user access pattern, request rate and storage and processing capacities. The proposed data replication scheme maximizes the availability, reliability and minimizes the access latency. Thus the proposed adaptive dynamic replication strategy improves the overall system performance and dynamically balances the load on addition and/or deletion of the nodes in the system. We have evaluated and compared the proposed replication method with existing data replication strategies. The simulation result shows that the proposed strategy results in better performance of the distributed system.
Volume 12 | 03-Special Issue
Pages: 1034-1043
DOI: 10.5373/JARDCS/V12SP3/20201349