Update shuffle documentation for branch-21.06 and UCX 1.10.1 #2475
Conversation
Signed-off-by: Alessandro Bellina <abellina@nvidia.com>
abellina
added
shuffle
|
@petro-rudenko, @yosefe fyi. |
|
build |
|
|
||
| 3. Fetch and install the UCX package for your OS and CUDA version | ||
| [UCX 1.9.0](https://github.com/openucx/ucx/releases/tag/v1.9.0). | ||
| UCX versions 1.10.1 requires the user to install `libnuma1`. RDMA packages have extra |
There was a problem hiding this comment.
Should we provide any instructions on installing libnuma1, or is it obvious enough that people need to an apt-get or yum install? Will this requirement go away in a future version of UCX?
There was a problem hiding this comment.
It'll install as package dependency: https://github.com/openucx/ucx/blob/v1.10.x/debian/control.in#L17
There was a problem hiding this comment.
It'll be installed as package dependency: https://github.com/openucx/ucx/blob/v1.10.x/debian/control.in#L17
There was a problem hiding this comment.
@petro-rudenko is right. For example: sudo apt install ./ucx-v1.10.1-ubuntu18.04-mofed5.x-cuda11.2.deb
The following NEW packages will be installed:
libnuma1 ucx
0 upgraded, 2 newly installed, 0 to remove and 420 not upgraded.
If you used dpkg -i the package is not automatically installed (but a apt install -f later does resolve). I'll remove this from the doc, as the package already specifies this.
There was a problem hiding this comment.
Changed our sample dockerfiles to use apt install [ucx-package].deb. I assume the same holds for CentOS, but we currently don't have a Dockerfile example for CentOS. I'll add, but maybe I should split these into dockerfiles rather than have them in the markdown.
There was a problem hiding this comment.
Ok trying a few things for CentOS, I'll add as a follow up issue and stick with Ubuntu (rdma-core is giving me trouble for CentOS).
|
@jlowe tweaked a few things and added a follow up, should be ready for another look. |
|
build |
…2475) * Update rapids-shuffle.md for UCX 1.10.1 Signed-off-by: Alessandro Bellina <abellina@nvidia.com> * Add message around JUCX 1.11.0 compatibility warning whenusing with UCX 1.10 * Update minimum requirement. JUCX 1.11.0 requires UCX 1.10+ * Small tweaks * Remove bullet points * libnuma1 pulled automatically from apt install
…2475) * Update rapids-shuffle.md for UCX 1.10.1 Signed-off-by: Alessandro Bellina <abellina@nvidia.com> * Add message around JUCX 1.11.0 compatibility warning whenusing with UCX 1.10 * Update minimum requirement. JUCX 1.11.0 requires UCX 1.10+ * Small tweaks * Remove bullet points * libnuma1 pulled automatically from apt install
This updates the documentation for the RAPIDS Shuffle Manager for the change to JUCX 1.11.0, and UCX 1.10.1+.
Closes #2286.