Knowledge check

  • 10 minutes

One of the R&D team's tightly coupled HPC applications is running more slowly than expected. It's a highly memory-bound, hybrid parallel application. Your team is running it on 16 HB120_v2 VMs, by using 30 MPI processes per VM and 3 threads per MPI process.

Based on this scenario, choose the best response for each of the following questions, then select Check your answers.

Check your knowledge

1.

You're using the Linux uptime tool to verify that the correct number of parallel processes and threads are running. What's the expected load average for this application on each VM?

2.

You're using the Linux top tool (with NUMA domain view, type 2) to verify that all parallel processes are distributed evenly among the NUMA domains. What percentage of user utilization is expected for each NUMA node domain?