VISION operator notices document service disruption and forthcoming maintenance

The customer problem is protecting research deadlines when infrastructure becomes unavailable. Include research-computing leadership, principal investigators, storage support and procurement. Ask which jobs can restart, what downtime the research calendar permits, and who owns vendor escalation. A bounded resilience assessment could map dependencies and rehearse one interrupted workload. The value hypothesis is reducing avoidable disruption and improving recovery predictability, subject to testing. Use the maintenance history to ask operational questions, not to estimate an annual failure rate or suggest that GPU hardware caused every incident. Do not promise uninterrupted operation or infer data loss from this log.

Build a dependency map covering scheduling, login services, network fabrics, storage and cooling. Establish a pre-change baseline for directory creation, rename and deletion alongside data-transfer throughput. Rehearse a storage change on an isolated representative environment and verify job restart behavior. Require approved test data, rollback artifacts and vendor support contacts. Validate permissions and output integrity after recovery. The proof of value should expose whether an apparently healthy bandwidth test misses a blocked application workflow. Evaluate representative jobs rather than only a device diagnostic; document which failure modes cannot safely be reproduced.

Assign the HPC operations lead as incident coordinator, with storage, facilities and security owners for their dependencies. Implement maintenance communications, checkpoint guidance, change rehearsals and post-change checks. Train users to distinguish paused access from failed jobs and to preserve recoverable work. Governance gates should cover change approval, rollback decisions and reopening. Proposed acceptance criteria are successful metadata operations within the locally agreed threshold, verified representative job resumption and complete notification records. Preserve root-cause uncertainty until the responsible team confirms it. Risks include incomplete vendor coordination, missed research deadlines and reopening before application-level validation.