Input staging, checkpoints, restart files and analysis all compete for storage. Capacity, bandwidth and metadata operations answer different questions.
Specify file counts, access patterns, concurrency and retention. A checkpoint test should include both writing and a successful restart; also account for moving the final results to the people who use them.
Evidence to bring
- Working set versus retained archive
- Checkpoint size and recovery window
- Shared access, metadata and data ownership
Record the test conditions and the person responsible for validating the result. Resolve the largest uncertainty before treating this as a deployable design.
PRIMARY SOURCES · CHECKED 9 NOVEMBER 2026
Dell: Validated Design for Government →2026-10-19 · Reference design applies to described versions and configurations; assess current applicability.
NERSC: job best practices →Living documentation; checked 2026-11-09 · Facility-specific operational guidance; use local site policies for actual jobs.