Dynamic Data Replication for Short Time-to-Completion in a Data Grid

Science collaborations use computer grids to run expensive computational tasks on large data sets. Tasks as jobs across the network demand data and thereby workload management and data allocation to maintain the computational workflow. Data allocation includes data placement with different replication factors (multiplicity) of data. The proposed data replication & allocation model can place multitudes of subsets of a data population in a distributed system, such as a computer cluster or computer grid. A stochastic simulation with a data and computing example from the ATLAS Physics Collaboration shows its potential usability in one of the largest Computing Grids. This paper showcases data allocation with different replica factors and various numbers of subsets to improve the overall situation in a computer network.