search

c3sr / comm_scope

NUMA-aware multi-CPU multi-GPU data transfer benchmarks
https://github.com/c3sr/scope
Apache License 2.0
21 stars 3 forks source link

NUMA node boundary benchmarks? #41

Closed rlerdorf closed 3 years ago

rlerdorf commented 3 years ago

Would it be hard to add benchmark tests that cross NUMA node boundaries? Perhaps by creating a shared memory segment from one NUMA node then measure latency and throughput reading and writing to it from another.

cwpearson commented 3 years ago

Hi @rlerdorf ,

These benchmarks just have an emphasis on CPU-GPU right now, so I wasn't planning on adding that. If you're interested in CPU-GPU or GPU-GPU across a NUMA node, those are already implemented though!

Carl

rlerdorf commented 3 years ago

Yeah, no problem. After a bit more reading I ended up writing my own cpu-cpu libnuma-based benchmark which is quite a bit simpler than what you are solving here.