namliz
commented
7 years ago Additional reference:
https://aws.amazon.com/ec2/virtualcores/ Kuberentes Docs: Managing Compute Resources
Meaning of CPU
Limits and requests for cpu are measured in cpus. One cpu, in Kubernetes, is equivalent to: 1 AWS vCPU 1 GCP Core 1 Azure vCore 1 Hyperthread on a bare-metal Intel processor with Hyperthreading
The simple stratagem:
- Instance type with more than 1 vCPU
- Boinc and distcc are deployed as daemonsets conditional on a node selector
- Tag nodes to match only one of these ahead of time (which one depends on the demo flags)
That's it. A full core per node is dedicated to boinc/distcc, to switch, simply edit the node labels ad watch as the controller does the right thing (destroys the no longer matching daemoset pods, create the newly matching ones).
Since its possible to access the kubernetes API from a pod the distcc sidecar that polls for new commits can also do the node relabeling to promote distcc.
This is a fair benchmark overall, it sets the floor at atleast 50% cluster utilization but I'm not yet entirely sure how excess CPU resources will get distributed.
Background reading:
https://github.com/kubernetes/kubernetes/blob/master/docs/devel/scheduler_algorithm.md https://github.com/kubernetes/kubernetes/blob/master/docs/design/resource-qos.md
This issue is to discuss what were dealing with in the context of the demo.