Methodology and Tools for Compute Performance at Any Scale
Clusters and Workstations are usually sized according to one-off benchmarking campaigns during tenders and way before production. While this widely used method is efficient, it has a number of drawbacks: (1) the benchmark data sets hardly capture day-to-day use cases in production; (2) this method assumes that Cluster’s Health and Performance are perfectly known and remain nominal anytime in production. Therefore it is usually really difficult to make sure that a user can maximize compute performance for a given job and software license budget.
This webinar will expose HP’s benchmarking method and show how the principle and benefits of this method can be leveraged in a day-to-day production. It will highlight some unique Intellectual property namely HP iLO4 * and HP CMU ** to perform “smart job scheduling”.