CAP : Computer-Aided Parallelization Framework for Multiprocessor Multidisk Storage Servers

The ability to execute remote operations on a storage server promises to reduce the bandwidth requirements between storage servers and clients. This contribution proposes a computer-aided parallelization (CAP) framework for creating parallel applications running on a multiprocessor multidisk storage server. Parallel operations are specified as a schedule of sequential operations with clearly defined inputs and outputs. The CAP specification of a parallel application consists of (1) the list of threads running on the storage server, (2) a set of sequential operations, (3) a schedule of sequential operations for each parallel operation, and (4) an assignment of sequential operations to threads. The schedule specification semantics is based on hierarchical directed acyclic graphs (HDAGs). This contribution shows how hierarchical directed acyclic graphs are translated into parallel programs, and how new operations (both sequential and parallel) can be dynamically added to an existing parallel program.