Dynamic Replication Management Scheme for Effective Cloud Storage

Abstract

Replication plays an important role for storage system to improve data availability, throughput and response time for user and control storage cost. Due to different nature of data access pattern, data popularity is important in replication because of the unstable and unpredictable nature of popular files. In addition, the replica placement is important in consideration of system's performance. In dataparallel applications, data locality is a key issue and the consequence of this issue occurs the decrement of system’ performance. Therefore, this thesis proposes a dynamic replication management scheme for effective cloud storage (ECS). The system contains two portions; replica allocation and replica placement. In the first portion, replica allocation, popularity is taken into account by analyzing the changes in data access pattern. Second for replica placement, replicas are placed and performed on dedicated assigned nodes in order to enhance data locality. The proposed placement algorithm is able to avoid the overloaded problem of nodes and the more effective storage utilization by considering the load of nodes; that is, disk utilization, CPU utilization and adjustable disk bandwidth. The proposed system is implemented as cloud storage system by using opensource CloudSim simulator. The aims of the proposed scheme are (i) to reduce unnecessary replication cost for unpopular data (ii) to achieve load balancing in data placement and (iii) to increase replica number for popular data. The analysis results demonstrated that the proposed scheme can adapt the degree of replication based on data popularity, save storage cost for unpopular files and achieve more load balancing than existing replication strategy such as Latest Access Largest Weight (LALW) algorithm.