The current template for the Linux "high mem" agents on ci.jenkins.io are currently of type "M5ad4xlarge" (check https://github.com/jenkins-infra/jenkins-infra/blob/production/dist/profile/templates/azure.ci.jenkins.io/agents.yaml.erb#L301-L338) .
These AWS instances types (as per https://aws.amazon.com/ec2/pricing/on-demand/) cost $0.824 per minute (on us-east-2), for 16 vCPUS (AMD EPYC clocked at 2,5 Ghz) 64 GiB, 2 x 300 NVMe SSD and network up to 10 Gigabit.
The main goal of these images is to provide a LOT of memory + clocked CPU to ensure that acceptance tests and heavy builds are used.
This task is about:
- Switching the instance size to a less costly one, with same amount of CPU/RAM
- Improving the usage of these machines: use excluvise labels and avoid reusing as their labels can be used to pick a "low" build, cosnuming time on such costy instance