Self-Hosting Runners

A Daisytuner runner - similar to for example github runners - exists to execute tests, performance evaluation workloads and with daisytuner, autotuning workloads.

Runners are grouped into partitions of specific available hardware. All runners in a partition are expected to support the same features / hardware interchangibly and will work through the queue of outstanding tasks of the partition.

If you need specialized hardware, don’t want your code running on our runners or have the hardware available, you may provision your own runners in a customer-specific partition. Your own runner partitions are available by name, next to all the other public partitions in Daisytuner workflows.

Daisytuner runners also capture detailed performance metrics from the underlying hardware for accurate performance measurements and thus need to run directly on the underlying hardware, with one instance per machine and root access.

Runners run in Node.JS and internally use docker as the environment to run workflows. They rely on server-side configuration which hardware to make available (and how) inside the docker containers.