Data agnostic resource scheduling in the grid

摘要

Data grid architectures must confront the antipodal problems of data locality and global data coherence. Here, we examine a new approach to meeting these conflicting needs that leverages a storage wide area network, a clustered file system, a distributed block system, and a distributed resource scheduler. We examine multi-site concurrent I/O performance on an emulated three-site data grid with inter-site separations ranging from 1000 km-5000 km. We examine workloads that require all sites to interact with a single, globally coherent data image. In particular, all three sites perform concurrent write operations to shared files within a single file system.