Re:GRIDiT – Coordinating Distributed Update Transactions on Replicated Data in the Grid

The recent proliferation of Grid environments foreScience applications led to common computing infrastructureswith nearly unlimited storage capabilities. Interms of data management, the Grid allows keeping alarge number of replicas of data objects to allow for ahigh degree of availability, reliability and performance.Due to the particular characteristics of the Grid,especially due to the absence of a global coordinator,dealing with many updateable replicas per data objecturgently requires new protocols for the synchronizationof updates and their subsequent propagation. Currentlythere is no protocol which can be seamlessly applied toa data Grid environment without impacting correctnessand/or overall performance. In this paper we addressthe problem of replication in the Data Grid in thepresence of updates. We have designed the Re:GRIDiTprotocol that focuses on the correct synchronization ofupdates to several replicas in the Grid in a completelydistributed way, extending well-established databasereplication techniques. Globally correct execution isprovided by communication between transactions andsites. Re:GRIDiT takes into account the special characteristicsof eScience applications such as the distinctionbetween mutable objects, that can be updated byusers and immutable objects. Finally, we provide a detailedevaluation of the performance of the Re:GRIDiTprotocol when being applied at Grid scale.