[Ofmfwg] Sunfish meeting notes for April 13, 2026

[Lee, Peter]

  *   Datacenter use case
     *   Discuss what needs to be exposed from the rabbits and what knapsack scheduler/Swordfish resource manager need to know to assign the right burst buffers to the hosts for a given workload. Leaning towards the abstraction of treating each rabbit as a self-contained NVMe sub-fabric. Knapsack should own physical topology awareness, scheduling, IP assignment and proximity decisions using the Cray System Manager’s global map, while Sunfish/Swordfish focus on logical Redfish/Swordfish objects and “same‑chassis” relationships.
     *   Will work on Friday towards a proposal for a rabbits based national labs mockup that is consistent with the flow diagram so we can work on a demo. (Latest version in the attached deck.)
     *   Russ pointed out there are gaps in current DMTF Redfish CXL modeling for FAM, where official mockup assume pre‑existing logical devices and fail to represent physical fabric adapters. This limits Sunfish’s ability to discover physical memory and endpoint topology before configuration.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openfabrics.org/pipermail/ofmfwg/attachments/20260413/aed56c16/attachment-0001.htm>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: Flux_Sunfish_NVMe_Discovery_v3.pptx
Type: application/vnd.openxmlformats-officedocument.presentationml.presentation
Size: 440871 bytes
Desc: Flux_Sunfish_NVMe_Discovery_v3.pptx
URL: <http://lists.openfabrics.org/pipermail/ofmfwg/attachments/20260413/aed56c16/attachment-0001.pptx>