Most queueing models assume that each job runs on a single server. But this one-server-per-job model is not a good representation of today’s compute jobs, particularly Machine Learning jobs. A typical data center job today occupies multiple cores concurrently, often thousands of cores. We refer to a job that ...
