Skip to content

wfaderhold21/benchmarks

Repository files navigation

benchmarks

Personal collection of HPC communication benchmarks covering SHMEM, MPI, UCX, UCC, and SHARP.

Actively maintained suites are marked with active; others are preserved for historical reference. This repository informs upstream discussions — see openucx/ucc PRs for related work.

Layout

Dir What it measures Build requirements
ucc/ active UCC collectives: alltoall, alltoallv (fixed + variable), pipelined a2av bandwidth, context/team creation timing MPI, UCC, UCX
ucx/ UCX-level alltoall and full HPCAC exchange patterns UCX, MPI
mpi/ RDMA all-to-all variants, MPI alltoall benchmark, scalability analysis MPI (RDMA-capable)
shmem/ SHMEM put/stencil/nonblocking/cuda benchmarks; Graph500; comprehensive test suites SHMEM (SYCL/MPI-based)
sharp/ Mellanox SHARP locality benchmark for in-network reduction SHARP, MPI
stream/ Endless copy/stream throughput micro-benchmark none
archive/ Historical results (e.g. 2019 SHMEM alltoall experiments)

Quick start

# UCC alltoall
cd ucc/a2a && make && mpirun -np 4 ./ucc_bench_a2a

# UCC context/team timing
cd ucc/timing-ucc && make && mpirun -np 4 ./time-ucc

# UCX-level benchmarks
cd ucx/alltoall && make && mpirun -np 4 ./alltoall

# MPI RDMA alltoall
cd mpi && make && mpirun -np 4 ./rdma_alltoall_demo

Output format

All UCC benchmarks emit CSV — see docs/output-format.md.

Results

Historical results under archive/.

About

No description, website, or topics provided.

Resources

Stars

1 star

Watchers

0 watching

Forks

Releases

No releases published

Packages

 
 
 

Contributors