Dynamic Checkpoint Data Replication Strategy in Computational Grid

Abstract

Computational grid is a good solution to large scale data processing and management problems including efficient checkpoint transfer and replication. Due to heterogeneous nature of grids most of time grids more prone to failure or latency delay. Subsequently checkpointing and replication is indispensable to tolerate such faults efficiently. Dynamic checkpoint data replication in computational grid aims to improve data access time and to utilize network and storage resources efficiently. Since the data checkpoints are very large and grid storages are limited, managing replicas in storage for the purpose of more effective utilization of them require more attention. In this work, a dynamic checkpoint data replication mechanism is proposed, which is called checkpoint based optimal replication (CBOR). CBOR selects a checkpoint for replication and calculates a suitable number of copies and grid sites for replication by setting different weight for each data access record. The data access records in the near past have higher weights. A grid simulator Optorsim is used to evaluate the performance of CBOR dynamic replication strategy. The experimental results show that CBOR successfully increases the effective network usage by finding out a popular checkpoint and replicates it to a suitable site.