While we’re always on the lookout for HPC news, not everything makes it to the front page. Notable items from this week include new benchmarks for 100GB Ethernet, the Open MPI roadmap, and Docker containers.
“Based on a containerized HPC environment this talk shows of a state-of-the-art stack including performance monitoring, log event handling and GraphDB based inventory to provide insights into what is going on within a SLURM cluster. The framework used is QNIBTerminal incorporating the ELK stack, a graphite backend and neo4j as a GraphDB.”
“100% Flash in the Datacenter? It won’t happen any time soon. Many (most?) tier one workloads will be moved to flash of course, but data is adding up so quickly that it’s highly unlikely you will be seeing a 100% datacenter any time soon. It will take a few years to have about 10/20% of data stored on flash and the rest will remain on huge hard disks (cheap 10+TB hard disks will soon be broadly available for example).”
In this slidecast, Christian Kniep presents: QNIBTerminal Plus InfiniBand – Containerized MPI Workloads. “QNIB Solutions (early on called ‘QNIB Inc’) derives from the first project Christian did during his B.Sc. report, a InfiniBand monitoring suite. For the sake of the report is was named ‘OpenIBPM: Open Source InfiniBand Performance Monitoring’. Afterwards Christian renamed it to match his last name (Kniep): ‘QNIB: Qualified Networking with InfiniBand’. Since then QNIB becomes a pet project’s theme.”