Shared resource among vma clients

Question

Shared resource among vma clients

ps142020 opened this issue 3 years ago · 2 comments

ps142020 commented 3 years ago

I am running an application with 12 of them launched on 12 different cores vs 1 core. I expected when all 12 process run on single core latency should increase due to obvious but turns out vma performs much better than 12 cores case. What is the shared resource here? i am not even running vma. Also from mlx_fs_dump i always see 12 queues. Is there any info on details of vma architecture?

Answer 1 · 2021-07-08T11:19:55.000Z

Any update on this?

Answer 2 · 2021-07-20T09:20:13.000Z

Hello @ps142020. You can find information about VMA using wiki: https://github.com/Mellanox/libvma/wiki or User Manual: https://docs.mellanox.com/spaces/viewspace.action?key=VMAv922