Post by nschellenberger on Jul 24, 2023 22:04:18 GMT
We have a PCIe card which acts as a FabricAdapter which can be placed in either an expansion Chassis (to expose the various other resources in that Chassis); and also in ComputerSystem to allow that host to consume resources from the various expansion Chassis. A CompositionService orchestrates the composition of the former into the latter.
Are we correct in understanding that:
The FabricAdapter when in a Chassis:
Conversely, for a FabricAdapter in a ComputerSystem:
An expansion Chassis:
A host ComputerSystem:
A host ComputerSystem's Chassis:
A PCIeDevice:
Even after eliminating the obviously inapplicable ones, there are so many "possible" relationships in the generic DMTF schemata that our UML diagrams look like an explosion in a yarn factory....
Are we correct in understanding that:
The FabricAdapter when in a Chassis:
- has ResourceBlocks for the exposed resources
- has e.g. Processors for any exposed processors
- has PCIeDevices for only its own device/functions (and not those of the resources exposed)
- has PCIeInterface to describe itself
Conversely, for a FabricAdapter in a ComputerSystem:
- has no ResourceBlocks, because it exposes nothing, acting only to attach the host to the fabric
- has PCIeDevices for only the adapter's own device/functions (and not those of the ComputerSystem itself)
- has PCIeInterface to describe itself
An expansion Chassis:
- has FabricAdapters pointing to the installed adapters
- has PCIeDevices pointing to all of the devices in the chassis (including the adapters)
- has PCIeSlots describing all of the slots in the Chassis (or has PCIeSlots been entirely superseded by Location/LocationType=Slot? or do we need both?)
A host ComputerSystem:
- has ResourceBlocks (one to advertise itself as an extensible system; and for each of the blocks composed from expansion chassis)
- empty Links/ConsumingComputerSystems and Links/SupplyingComputerSystems (since we don't permit composition of host systems, only from expansion chassis)
- has Processors for all processors, both local and remotely composed
- has a Chassis for its own enclosure (in our case really only because we need to model the PCIeDevices which are ContainsTarget=true only on Chassis)
A host ComputerSystem's Chassis:
- empty ResourceBlocks
- empty FabricAdapters
- has PCIeDevices for all devices both local and remotely composed
A PCIeDevice:
- has PCIeDevice/PCIeInterface describing itself
- has PCIeDevice/Slot describing itself, also (why is this separate from PCIeInterface?)
Even after eliminating the obviously inapplicable ones, there are so many "possible" relationships in the generic DMTF schemata that our UML diagrams look like an explosion in a yarn factory....