Question 270
A team has corrected this HPE compute solution issue: A team connects GPU training nodes through one undersized shared link and sees long distributed job delays. Which continuing practice would best prevent the same problem?
Explanation
Correct Answer
Size interconnect capacity using distributed job measurements. This recurring assurance measure is tied to the diagnosed risk: Inter-node communication bandwidth is mismatched to model parallelism and data movement. For the active incident, the corrective action remains: Validate the supported high-bandwidth network topology for the planned distributed workload.
Incorrect Answers
Inter-node communication bandwidth is mismatched to model parallelism and data movement. Using the proposed safeguard would guard against a separate risk: The thermal design lacks the required resilience against a planned cooling interruption. Size interconnect capacity using distributed job measurements.
This control protects against a condition in which accelerator memory has not been sized for concurrent resident models and serving overhead. The issue to prevent here is: Inter-node communication bandwidth is mismatched to model parallelism and data movement. Size interconnect capacity using distributed job measurements.
Inter-node communication bandwidth is mismatched to model parallelism and data movement. The proposed safeguard is intended for another failure mechanism: Model load time and repeated I/O exceed the available storage path performance. Size interconnect capacity using distributed job measurements.