-
Notifications
You must be signed in to change notification settings - Fork 8
Open
Description
What problem are you trying to solve?
General architecture
Describe the solution you’d like
- quantum resources must be able to be on-node and off-node
- agnosticity
- q: how devices topology is exposed? topology is not needed for wms, but for end user/scientist it is a requirement to know. Consideration: what is the minimum criteria that user can specify?
- for admin: quantum systems might need some kind of parallelism abstract (execution lanes of IBM, middleware for Pasqal, etc.). How this abstract is mapped to resource in slurm? How user defines parallelism (estimation of parallel execution; capacity planning: timeslots, credits)?
- for the admin: ability to decide how sharable systems are (ex: exclusive, non-exclusive access). Ex of implementation: licences.
- Jobs can request hardware to have exclusive access.
- quantum resources should be configurable? Let's have docs on what is qpu, quantum system, qubit, etc.?
- Quantum resources must exposed metrics and accounting information (utilization, power draw, etc.). Quantum device should expose it and Slurm should be able to consume it.
- resource api are separated into 3 buckets: resource acquisition, job running and accounting/metrics.
- versioning. Backward compatibility is required (how would we achieve this?)
Describe any alternatives you’ve considered
No response
Documentation, Adoption, Migration Strategy
No response
Metadata
Metadata
Assignees
Labels
No labels