A Look at Communication-Intensive Performance in Julia

The Julia programming language continues to gain popularity both for its potential for programmer productivity and for its impressive performance on scientific code. It thus holds potential for large-scale HPC, but we have not yet seen this potential fully realized. While Julia certainly has the machinery to run at scale, and while others have done so for embarrassingly parallel workloads, we have yet to see an analysis of Julia's performance on communication-intensive codes that are so common in the HPC domain. In this paper we investigate Julia's performance in this light, first with a suite of microbenchmarks within and without the node, and then using the first Julia port of a standard, HPC benchmarking code, high-performance conjugate gradient (HPCG). We show that if programmers properly balance the computation to communication ratio, Julia can actually outperform C/MPI in a cluster computing environment.