Trying to use frequency governor on the utah cluster

[Alan Zhu]

We're trying to do some cluster/resource management experiments which require scaling the frequency of certain cores, using cpupower. The only cpufreq drivers I see when I run ls /usr/lib/modules/$(uname -r)/kernel/drivers/cpufreq/ are:

amd_freq_sensitivity.ko  p4-clockmod.ko  speedstep-lib.ko

when I think we need something like amd_pstate.ko. I'm wondering if this is a BIOS setting or something that we need to toggle - alternatively, is there a way to do frequency scaling that I'm missing?

Experiment name: vedantha-206644

Thanks,

Alan

[ajma...@gmail.com]

Hi Alan,

This does seem like a misconfigured BIOS, these nodes should all allow for frequency scaling and on amd211 I see "no or unknown cpufreq driver is active on this CPU".  On another node I have allocated I see the following:

# cpupower frequency-info
analyzing CPU 0:
  driver: acpi-cpufreq
  CPUs which run at the same hardware frequency: 0
  CPUs which need to have their frequency coordinated by software: 0
  maximum transition latency:  Cannot determine or is not supported.
  hardware limits: 1.50 GHz - 3.00 GHz
  available frequency steps:  3.00 GHz, 2.40 GHz, 1.50 GHz
  available cpufreq governors: conservative ondemand userspace powersave performance schedutil

...

The node I ran that command on doesn't have `amd_pstate.ko` either, but it still allows for performance governance, so that specific module may not be needed unless there's a specific reason you need to use that exactly.

Is it alright if I restart your nodes to make BIOS changes?

Best,

 - Aleks

[ajma...@gmail.com]

I found the offending BIOS setting via racadm, and changed it to what it's supposed to be for both of your nodes.  The change is staged and should apply when you next reboot your nodes, you can do so at your own convenience.  Let me know if that helps.

Best,

 - Aleks