Grid Data Management Pilot (GDMP): A Tool for Wide Area Replication

The CMS experiment at CERN, the European Organization for Nuclear Research, is currently setting up a Data Grid which will provide access to several Terabytes of data, distributed and replicated. In the real production environment data will be produced in different countries on both sides of the Atlantic by the end of year 2000. The stringent requirements of data consistency, security and high-speed transfer of huge amounts of data, imposed by the physics community need to be satisfied by an asynchronous replication mechanism. A pilot project called the Grid Data Management Pilot (GDMP) has been initiated which is responsible for asynchronously replicating large object-oriented data stores over the wide-area network to globally distributed sites. We present the design, architecture, functionality and performance results of our first working prototype. Different replication policies and protocols are supported that range from strictly synchronous to rather relaxed asychnronous models in terms of data consistency. We believe that this first prototype can be regarded as a pioneer step towards a Data Grid and as a prototype for replication management within other Data Grid approaches like DataGrid, GriPhyN and PPDG [1, 2, 3]. keywords: Grid computing, data management, replication