Accelerator Plus includes a hierarchical scheduler architected to offload the base scheduler — enabling greater throughput, better license and resource utilization, and more flexible scheduler usage models — plus millisecond dispatch latency, which yields the most benefits for short jobs. It serves as a “personal scheduler” director, allowing users to queue short jobs sequentially on their own. This reduces the scheduling burden on the lower-level scheduler by presenting a batch of short jobs as one longer job while maintaining full visibility into each individual job. It can also handle all user query, job submission, and reporting functions. The offload from the base scheduler is substantial, since 80% of typical scheduler loads come from these non-dispatch functions.
For short, highly parallel jobs with consistent license and computing resource needs, Accelerator Plus deployments have realized a 6-10X increase in throughput with no added compute capacity.
Accelerator Plus is equipped with a comprehensive set of policy management features including FairShare and preemption. Cross-organizational resources can be assigned and managed as users prefer – by workflow, by project, or on demand. A full-featured GUI enables full visibility into each job and compute farm statistics for queued jobs, running jobs, and allocated resources. User interfaces include web browsers, command line, and an extension API.
For example, in a semiconductor verification application, under a typical high-load situation of two regressions and a smoke test, Accelerator Plus manages the peak loads with ease. When combining all three workloads, the effective load on the base scheduler stays stable at a level that is less than any of the individual workloads.