Tuning MPI Collectives by Verifying Performance Guidelines

ABSTRACT MPI collective operations provide a standardized interface for performing data movements within a group of processes. The e ciency of collective communication operations depends on the actual algorithm, its implementation, and the speci c communication problem (type of communication, message size, number of processes). Many MPI libraries provide numerous algorithms for speci c collective operations. The strategy for selecting an e cient algorithm is often times prede ned (hard-coded) in MPI libraries, but some of them, such as Open MPI, allow users to change the algorithm manually. Finding the best algorithm for each case is a hard problem, and several approaches to tune these algorithmic parameters have been proposed. We use an orthogonal approach to the parameter-tuning of MPI collectives, that is, instead of testing individual algorithmic choices provided by an MPI library, we compare the latency of a speci c MPI collective operation to the latency of semantically equivalent functions, which we call the mock-up implementations. The structure of the mock-up implementations is de ned by selfconsistent performance guidelines. The advantage of this approach is that tuning using mock-up implementations is always possible, whether or not an MPI library allows users to select a speci c algorithm at run-time. We implement this concept in a library called PGMPITuneLib, which is layered between the user code and the actual MPI implementation. This library selects the best-performing algorithmic pattern of an MPI collective by intercepting MPI calls and redirecting them to our mock-up implementations. Experimental results show that PGMPITuneLib can signi cantly reduce the latency of MPI collectives, and also equally important, that it can help identifying the tuning potential of MPI libraries.