GoogleCloudPlatform / GoogleCloudPlatform/compute-virtual-ethernet-linux

`ethtool --set-rxfh-indir` does not work on H100 instances (`GVE_DEV_OPT_ID_RSS_CONFIG` is not given?)

Open
#69 6 comments 0 reactions 0 assignees View on GitHub
Dominant language
C
Stars
91
Forks
44
Avg merge
1h 21m
Merged PRs (30d)
2

Description

I'm unable to run commands like

```
ethtool --set-rxfh-indir eth1 equal 16 # Output: Cannot get RX flow hash indirection table size: Operation not supported
ethtool -x eth1 # Also doesn't work
```

when building gVNIC v1.4.5 from source on both a3-highgpu and a3-megagpu instances. The former command is necessary to get [tcpgpudmarxd](https://github.com/google/tcpgpudmarxd) working for multi-node training.

I believe this is because the device option `GVE_DEV_OPT_ID_RSS_CONFIG` is not given to the driver (i.e. [this code](https://github.com/GoogleCloudPlatform/compute-virtual-ethernet-linux/blob/d5ad781d87806deb507a9760d6b11a9ab4144eca/google/gve/gve_adminq.c#L211) is never run). There is an unrecognized option `0x9` but it doesn't seem to be for RSS.

I'm running on `6.10.0-2.el9ueknext.x86_64`.

These commands do work in Container Optimized OS 105, but the gVNIC driver included with that operating system does not appear to be the same as the gVNIC driver in this repository (compare the ethtool implementations: [COS](https://cos.googlesource.com/third_party/kernel/+/cc53f55433a4e01ea51d28dde1594514252eace2/drivers/net/ethernet/google/gve/gve_ethtool.c#685), [this repo](https://github.com/GoogleCloudPlatform/compute-virtual-ethernet-linux/blob/d5ad781d87806deb507a9760d6b11a9ab4144eca/google/gve/gve_ethtool.c#L844)).

How can I get `ethtool --set-rxfh-indir` working on a3-highgpu instances and a3-megagpu instances with gVNIC built from source? Thank you for your help!

Contributor guide

Open the contributing guide

Assessment

This issue has not been assessed yet.

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.