[
https://issues.apache.org/jira/browse/YARN-9120?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16722122#comment-16722122
]
Peter Bacsko commented on YARN-9120:
------------------------------------
{quote}
I'm also thinking in what scenario would a user wants a diverse configuration
in the cluster. As far as I know, the Ambari can only update configuration at
the cluster level.
{quote}
IMO it's very reasonable to assume that GPU-based cluster will not homogeneous.
If you think about it, an Nvidia V100 costs around 5000-6000 USD which is a lot
of money. If a cluster consists of 500 nodes, it's very likely that they
install only 20-30 of these silicone beasts (just a random number). Equipping
every node with an expensive GPU is a really big investment.
As far as the new property is concerned, the new value {{off}} makes sense to
me, it looks more natural.
> Need to have a way to turn off GPU auto-discovery in GpuDiscoverer
> ------------------------------------------------------------------
>
> Key: YARN-9120
> URL: https://issues.apache.org/jira/browse/YARN-9120
> Project: Hadoop YARN
> Issue Type: Improvement
> Reporter: Szilard Nemeth
> Assignee: Szilard Nemeth
> Priority: Major
>
> GpuDiscoverer.getGpusUsableByYarn either parses the user-defined GPU devices
> or should have the value 'auto' (from property:
> yarn.nodemanager.resource-plugins.gpu.allowed-gpu-devices)
> In some circumstances, users would want to exclude a node from scheduling, so
> they should have an option to turn off auto-discovery.
> It's straightforward that this is possible by removing the GPU
> resource-plugin from YARN's config along with GPU-related config in
> container-executor.cfg, but doing that with a dedicated value for
> yarn.nodemanager.resource-plugins.gpu.allowed-gpu-devices is a more
> lightweight approach.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
