Slurm and workload management
I configure shared clusters so researchers can submit work with fewer surprises, and help administrators explain why a job is waiting, failing or wasting resources.
Build scheduling policy around the research
A useful queue policy reflects the workloads and the people sharing the cluster. I work with your team on partitions or queues, fair share, backfill, limits, GPU and license scheduling, and usage accounting. The scope covers one agreed Slurm, PBS Pro or LSF scheduler environment.
We establish which resources a job needs together, which projects have priority and which limits must be enforced. That makes the policy easier to explain to users and gives administrators a clearer basis for diagnosing pending work.
Verify allocation and enforcement together
Correct scheduler records are only part of a working configuration. I test representative jobs against the resource setup, including CPU and memory behavior and GPU allocation where included in the environment. The checks distinguish what the scheduler reserves from what the operating system and runtime enforce.
A software or compute-image update can change that relationship. Administrator handover includes the relevant checks and configuration record, so the team has a way to validate later changes instead of relying on a one-time successful launch.
Investigate difficult jobs
Job troubleshooting follows the effective request, step records, application output and node observations. Pending work may be blocked by resource shape, an account or QOS limit, a reservation or scheduling priority. Failed or stalled work may need memory, launcher, storage or communication investigation.
For GPU and distributed jobs, I check the supported launch and container path rather than assuming that a running process proves useful progress. Changes are tested with an agreed reproduction before being applied more widely.
Make the setup maintainable
The engagement includes setup, configuration, agreed tuning, representative workload testing, written recommendations and administrator handover. Scheduler migration can be discussed alongside the workflows and software that need to continue working through it.
You can book after a brief qualification call, without first buying an architecture assessment. Written scope confirms the system, tasks, workloads, access, testing window and delivery date. Hardware, software licenses, travel, ongoing support and work outside that scope are separate.
Engagement options
Choose a format below to see current rates and the ordering options for that engagement. If you are unsure which fits, we can work through the scope together.
- Scheduler setup and workload reviewA defined scheduler setup and workload review.