Load-Aware Dynamic Replication Management in a Data Grid

Data Grids are increasingly popular in novel, demanding and dataintensiveeScience applications. In these applications, vast amounts of data, generatedby specialized instruments, need to be collaboratively accessed, processedand analyzed by a large number of users spread across several organizations. Thenearly unlimited storage capabilities of Data Grids allow these data to be replicatedat different sites in order to guarantee a high degree of availability. Forupdateable data objects, several replicas per object need to be maintained in aneager way. In addition, read-only copies serve users’ needs of data with differentlevels of freshness. The number of updateable replicas has to be dynamicallyadapted to optimize the trade-off between synchronization overhead and the gainwhich can be achieved by balancing the load of update transactions. Due to theparticular characteristics of the Grid, especially due to the absence of a globalcoordinator, replication management needs to be provided in a completely distributedway. This includes the synchronization of concurrent updates as wellas the dynamic deployment and undeployment of replicas based on actual accesscharacteristics which might change over time. In this paper we present theRe:GRIDiT approach to dynamic replica deployment and undeployment in theGrid. Based on a combination of local load statistics, proximity and data accesspatterns, Re:GRIDiT dynamically adds new replicas or removes existing oneswithout impacting global correctness. In addition, we provide a detailed evaluationof the overall performance of the dynamic Re:GRIDiT protocol which showsincreased throughput with respect to the replication management protocol with astatic number of replicas.