Adding CCC based autoscaler files

[pulasthi]

Adding scripts for automated microbenchmark runs with CCC

[linamy85]

Thanks @pulasthi! One question: when running the automation_launch.sh, I can see a lot of warning as such

Warning: Key 'cloud.google.com/gke-tpu-topology' is not recommended with node selector; Consider using Custom Compute Classes mechanisms, simultaneous use of both may lead to unexpected behavior, use with caution.
Warning: Key 'cloud.google.com/gke-tpu-accelerator' is not recommended with node selector; Consider using Custom Compute Classes mechanisms, simultaneous use of both may lead to unexpected behavior, use with caution.

Should we remove them from yaml?

[junjieqian]

We also need to add a README and update https://github.com/AI-Hypercomputer/accelerator-microbenchmarks/blob/tpu7x-auto/Ironwood/guides/automation/README.md to include the ccc solution.
Thanks

[pulasthi]

Thanks @pulasthi! One question: when running the automation_launch.sh, I can see a lot of warning as such

Warning: Key 'cloud.google.com/gke-tpu-topology' is not recommended with node selector; Consider using Custom Compute Classes mechanisms, simultaneous use of both may lead to unexpected behavior, use with caution.
Warning: Key 'cloud.google.com/gke-tpu-accelerator' is not recommended with node selector; Consider using Custom Compute Classes mechanisms, simultaneous use of both may lead to unexpected behavior, use with caution.

Should we remove them from yaml?

This is a limitation in CCC when its been used in conjunction with Kueue. When CCC is used without Kueue the only nodeselector tag that is needed is the one pointing to the CCC class

    cloud.google.com/compute-class: tpuv7-2x2x1-class

However using just this Kueue does not create the workload correctly and hence the normal node selectors of

    cloud.google.com/compute-class: tpuv7-2x2x1-class
    cloud.google.com/gke-tpu-accelerator: tpu7x

need to be added to get pass Kueue. However CCC logs warnings for these since this information is already present in the CCC template. I will fill a bug about this and followup with the CCC team. But for now we would need to keep these

[pulasthi]

We also need to add a README and update https://github.com/AI-Hypercomputer/accelerator-microbenchmarks/blob/tpu7x-auto/Ironwood/guides/automation/README.md to include the ccc solution. Thanks

Hi @junjieqian added the updated readme file. Most of the content is same as the current readme file