Engineering Papers⌕ Search

Engineering topics

Traff, J L

Publications and source records attributed to Traff, J L.

A simple, pipelined algorithm for large, irregular all-gather problems.

We present and evaluate a new, simple, pipelined algorithm for large, irregular all-gather problems, useful for the implementation of the MPI-Allgatherv collective operation of MPI. The algorithm can be viewed as an adaptation of a linear ring algorithm for regular all-gather problems for single-ported, clustered multiprocessors to the irregular problem. Compared to the standard ring algorithm, whose performance is dominated by the largest data size broadcast by a process (times the number of processes), the performance of the new algorithm depends only on the total amount of data over all processes. The new algorithm has been implemented within different MPI libraries. Benchmark results on NEC SX-8, Linux clusters with InfiniBand and Gigabit Ethernet, Blue Gene/P, and SiCortex systems show huge performance gains in accordance with the expected behavior.

99 GENERAL AND MISCELLANEOUS↗

Self-consistent MPI-IO performance requirements and expectations.

We recently introduced the idea of self-consistent performance requirements for MPI communication. Such requirements provide a means to ensure consistent behavior of an MPI library, thereby ensuring a degree of performance portability by making it unnecessary for a user to perform implementation-dependent optimizations by hand. For the collective operations in particular, a large number of such rules could sensibly be formulated, without making hidden assumptions about the underlying communication system or otherwise constraining the MPI implementation. In this paper, we extend this idea to the realm of parallel I/O (MPI-IO), where the issues are far more subtle. In particular, it is not always possible to specify performance requirements without making assumptions about the implementation or without a priori knowledge of the I/O access pattern. For such cases, we introduce the notion of performance expectations, which specify the desired behavior for good implementations of MPI-IO. I/O performance requirements as well as expectations could be automatically checked by an appropriate benchmarking tool.

97 MATHEMATICS AND COMPUTING↗