問題文
A team wants a quick confirmation, as part of the per-server checks, that the collective communication path inside each machine works at all. They do not yet need a full performance characterization. Which approach fits this goal?
選択肢
- A path that works inside one server will necessarily work between servers as well and the per-server check adds nothing to what the burn-in will establish, so schedule a multi-node burn-in immediately.
- Use the collective test that the diagnostic suite includes, since it completes in a few minutes and confirms the path is functional.
- That is the only level that includes any collective communication check and a shorter run would skip the path the team wants to confirm, so run the longest available level of the diagnostic on each server.
- Write a custom program that exercises the path.