Platform Aware Scheduling v0.10
Pre-release
Pre-release
General
- Updated dependencies with Kubernetes 1.25 and Go 1.19
- Improvement to the descheduling & labelling strategies
- Security improvements to the TAS and custom-metrics-api containers and of the health-metric demo
- Address issues with Dockerfile and code license and copyright
- Documentation updates
- Linting improvements
Telemetry Aware Scheduling
- Release of telemetry-aware-scheduling/v0.4.0 aligned to the container image
- Improvements to the descheduling and labelling strategies:
-
- Strategies are enforced sequentially to reduce number of spawned goroutines: 75cf83e
-
- Update burst and QPS parameters for K8s go client to avoid throttling: 75cf83e
- Bump to K0s v1.25 and go 1.19: d00722e
- Improved security of the TAS and custom-metrics-api containers and of the health metric demo: 692d5ff
- Address issues with Dockerfile and code license and copyright: 5661d46, 6f6883f
GPU Aware Scheduling
- GPU allocations can be targeted to Xe Linked cards
- Support single-numa GPU allocations, when multiple GPUs are allocated.
- Bugfixes, especially around split (long) label handling