-
Feature Request
-
Resolution: Unresolved
-
Normal
-
None
-
None
-
None
-
Product / Portfolio Work
-
None
-
False
-
-
None
-
None
-
None
-
-
None
-
None
-
None
-
None
-
None
1. Proposed title of this feature request
"AMD Last Level Cache Affinity for MicroShift"
2. What is the nature and description of the request?
Performance sensitive workload running on AMD Epyc CPUS can suffer from noisy neibourhs and Last Level Cache (LLC) misalignment. A topology-aware NRI plugin policy can avoid mis-placement / misalignment of containers/pods. The goal is to have a supported way of configuring / using LLC affinity with MicroShift.
3. Why does the customer need this? (List the business requirements here)
Better support of performance sensitive workload.
4. List any affected packages or components.
MicroShift
More infos on AMD LLC affinity:
https://github.com/AMDEPYC/nri-plugins
A generic solution is being added to k8s via this KEP:
https://github.com/kubernetes/enhancements/tree/master/keps/sig-node/4800-cpumanager-split-uncorecache#design-details
Probably need only to ensure that this also works with MicroShift.
- is triggering
-
OCPSTRAT-2333 Last Level Cache locality for MicroShift
-
- Backlog
-
-
OCPSTRAT-2342 Improve FeatureGate handling in MicroShift
-
- Backlog
-