A Two-layer Geo-cloud based Dynamic Replica Creation Strategy

Abstract

To improve data availability and reduce user access latency, geo-cloud based data replication is widely used in large global Web sites, such as Facebook. However, as the popularity of data is different and will change as time goes by, simple static replica creation strategies that assign the same number of replicas to all data, nev er changing thereafter, are not suitable. To this issue, we propose a two-layer geo-cloud based dynamic replica creation strategy called TGstag. TGstag addresses the issue with twofold: policy constraint heuristic inter-datacenter replication and load aware adaptive intra-datacenter replication. TGstag aims to minimize both cross-datacenter bandwidth consumption and average access time with constraints of policy and commodity node capacity. To evaluate the effectiveness of our strategy, we've conducted comprehensiv e experiments to compare TGstag with other approaches. The results show TGstag significantly reduces cross-datacenter bandwidth and ave rage read access time.