compute_allocation How to optimally allocate compute resources such as OnDemand, Reserved VMs and Serverless for a user workload?