Daily Bulletin

Derecho weekly update

March 3, 2023

The NWSC-3 project team reached a significant milestone this week by completing work on hardware and software commissioning. HPE and CISL engineers began configuring and customizing Derecho to integrate it into NCAR’s HPC environment. HPE engineers also worked to resolve errors on the HPE Slingshot interconnect, and oversight monitoring of the fabric looked good. 

Benchmarking experts started running the HPE system checker to check the consistency of firmware (settings, BIOS, OS, SW stack, etc.) across systems, especially the compute blades and the interconnect fabric, and things looked good overall. HPE reports seeing consistent performance on Derecho.

Linpack load was used to stress the system so the NWSC team could monitor the facility’s power usage effectiveness (PUE) in real time, a capability the team recently implemented. They recorded a 1.9MW power draw on Derecho’s CPU nodes and a 4.2MW draw for the entire site, which included the Cheyenne system. During the Linpack runs the overall PUE was 1.07 -1.09. The facility handled the load well and no alarms for either the mechanical or electrical equipment were generated.

If all goes well with the NCAR integration work, system health checks, and HPE benchmarks, we will start acceptance testing in the next few days.