Optimizing bulk data transfers using network measurements: A practical case

In modern Data Grid infrastructures, we increasingly face the problem of providing the running applications with fast and reliable access to large data volumes, often geographically distributed across the network. As a direct consequence, the concept of replication has been adopted by the grid community to increase data availability and maximize job throughput. To be really effective, such process has to be driven by specific optimization strategies that define when and where replicas should be created or deleted on a per-site basis, and which replicas a job should use. These strategies have to take into account the available network bandwidth as a primary resource, prior to any consideration about storage or processing power. We present a novel replica management service, integrated within the Gluedomains active network monitoring architecture, designed and implemented within the centralized collective middleware framework of the SCoPE project to provide network-aware transfer services for data intensive Grid applications.