autosize control plane support for performance profile creator

[ffromani]

Implement reserved cpu (aka infra+control plane) sizing using a the linear programming optimization (gonum/optimize).

The core idea is to model the constraints and let the optimization package compute the desired target.

These changes where AI-Assisted (hence the AA tag), then largely amended by a human (hence the HI tag - Human Intervention).

The initial penalty cost structure was suggested by google Gemini 2.5 flash, and then amended by human intervention.

[openshift-ci]

[APPROVALNOTIFIER] This PR is APPROVED

This pull-request has been approved by: ffromani

The full list of commands accepted by this bot can be found here.

The pull request process is described here

Needs approval from an approver in each of these files:

• OWNERS [ffromani]

Approvers can indicate their approval by writing /approve in a comment
Approvers can cancel approval by writing /approve cancel in a comment

[ffromani]

/test e2e-no-cluster

[ffromani]

/cc @MarSik

[MarSik]

This is an interesting approach, I wonder if we can express constraints like allocate the whole (multiples) L3 CCDs for reserved or integrate the NIC queue count (no more than 16/32) and interrupt counts (224 per cpu).

[ffromani]

This is an interesting approach, I wonder if we can express constraints like allocate the whole (multiples) L3 CCDs for reserved or integrate the NIC queue count (no more than 16/32) and interrupt counts (224 per cpu).

We can introduce a quantitative limitation like I did for SMT, so we allocate in such a way to minimize the LLC count, for example. The problem however is that we only have the quantitative axis because kubelet owns the exact placement. We can say that, overall, 7 CPUs is better than 9 because we have all aligned and then we can make best use of compute resources (completely made up numbers, hope it is clear enough)

[ffromani]

typo: distinguish

[MarSik]

@ffromani Well the PPC can (and should) select the specific cpus for the reserved/isolated split based on the capacity computation. It has all the hardware topology information to be able to do that. Or are we talking about different aspects of PPC here?

[openshift-ci]

@ffromani: The following tests failed, say /retest to rerun all failed tests or /retest-required to rerun all mandatory failed tests:

Test name	Commit	Details	Required	Rerun command
ci/prow/lint	6005e98	link	true	/test lint
ci/prow/e2e-aws-ovn-techpreview	6005e98	link	true	/test e2e-aws-ovn-techpreview
ci/prow/e2e-aws-ovn	6005e98	link	true	/test e2e-aws-ovn
ci/prow/okd-scos-e2e-aws-ovn	6005e98	link	false	/test okd-scos-e2e-aws-ovn

Full PR test history. Your PR dashboard.

Instructions for interacting with me using PR comments are available here. If you have questions or suggestions related to my behavior, please file an issue against the kubernetes-sigs/prow repository. I understand the commands that are listed here.

[ffromani]

@ffromani Well the PPC can (and should) select the specific cpus for the reserved/isolated split based on the capacity computation. It has all the hardware topology information to be able to do that. Or are we talking about different aspects of PPC here?

Uhm, we can achieve that rethinking all the core allocation stage. I added an add-on step to demo the autosizing, but we are looking to a possible full rewrite of the allocation logic embedding some form of optimization, this is a much bigger endeavour. Do we want to go this direction?