Code

Vmm benchmark support#883

Code

Draft

pentschev wants to merge 15 commits intorapidsai:branch-0.28rapidsai/ucx-py:branch-0.28from pentschev:vmm-supportpentschev/ucx-py:vmm-supportCopy head branch name to clipboard

+354-73

Commits

Commits on Sep 28, 2022

• Add VMM allocator

  pentschev

  committed
  e7ad896

  Copy full SHA for e7ad896

• Use VMM allocator in send/recv benchmark

  pentschev

  committed
  a72af2e

  Copy full SHA for a72af2e

• Add VmmArray extension class for VmmAllocPool

  pentschev

  committed
  e903847

  Copy full SHA for e903847

• Support VmmArray extension in send/recv benchmark

  pentschev

  committed
  7f3800f

  Copy full SHA for 7f3800f

• Add --vmm-debug flag

  pentschev

  committed
  d6ee493

  Copy full SHA for d6ee493

• Remove old VMMArray implementation

  pentschev

  committed
  9f9dd98

  Copy full SHA for 9f9dd98

• Rename module vmm_ext to vmm

  pentschev

  committed
  105db68

  Copy full SHA for 105db68

Commits on Sep 30, 2022

• Add support for VmmBlockPool in send/recv benchmark

  pentschev

  committed
  fdf7885

  Copy full SHA for fdf7885

Commits on Oct 3, 2022

• Add support for VmmPool

  pentschev

  committed
  2a8c4dc

  Copy full SHA for 2a8c4dc

• Support VMM blocks in Array

  pentschev

  committed
  1d3aa83

  Copy full SHA for 1d3aa83

• Support VMM blocks in Endpoint send/recv methods

  pentschev

  committed
  be07fe3

  Copy full SHA for be07fe3

Commits on Oct 4, 2022

• Do not send entire block, send only the part used by allocation

  pentschev

  committed
  4c35dbe

  Copy full SHA for 4c35dbe

• Support RMM-backed VMM pool

  pentschev

  committed
  7df2c73

  Copy full SHA for 7df2c73

• Do not use asyncio.gather to guarantee message ordering

  pentschev

  committed
  3e187e3

  Copy full SHA for 3e187e3

Commits on Oct 5, 2022

• Ensure slice doesn't go beyond allocation size

  pentschev

  committed
  1d89f0e

  Copy full SHA for 1d89f0e