Open Stack

Hi, Vladimir/Mate/Yaguang
	I'm considering the PCI pass-through support before I noticed related blueprints listed below. I'm considering if we can combine the GPU pass-through and general PCI pass-through together, or even more, implement a generic direct hardware resource assignment support, so that we can even easily to extend to USB and other devices (hmm, will anyone want to provide USB in cloud environment? ). After all, they are all resource that can't be shared, thus need carefully inventory management, need present usage status to user, possibly need some setup/clean before assigned to a new guest, need be released explicitly after instance destroyed etc.

https://blueprints.launchpad.net/nova/+spec/xenapi-gpu-passthrough 
https://blueprints.launchpad.net/nova/+spec/pci-passthrough-and-sr-iov 
https://blueprints.launchpad.net/nova/+spec/pci-passthrough

	Below is some idea from my side, and hope to get your feedback.

	1) How the compute node present the hardware resources? Each compute node will configure host specific hardware resources information, including "resource type", "resource description", "resource information", "resource handler classpath", "resource count" etc. Below are examples for vGPU in xen API, SR-IOV network devices, some encryption devices:
		ResourceType: 			VGPU
		Resource Information:		PGPU grougp ID
		Resource Description:		XXXXXXX
		Resource Count:			2
		Resource handler class path:	nova.virt.xenapi.vgpu

		ResourceType:			NIC_VF_1G
		Resource Information:		PCI path: Bus:Device:vFunction
		Resource Description:		SR-IOV NIC with 1G bandwidth
		Resource Count:			3
		Resource handler class path: nova.virt.libvirt.pci

		ResourceType:			Encryption_card
		Resource Information:		USB: VendorID: ProductID
		Resource Description:		Encryption card to encrypt data
		Resource Count:			1
		Resource handler class path: nova.virt.libvirt.usb

	2) How the hardware resources managed globally? As Mate stated, each compute periodically report available hardware resources _publish_service_capabilities @ manager, the information should include [type, description, count] and is placed in host_state.capabilities. These information can be used by filter scheduler.

	Several details need finalized, like how to make sure the same resource type in different host has same description, how to place such information in the database, what's the path in host_state.capabilities etc. But that can be resolved during implementation stage.

	3) How user specifies a given instance needs hardware resource? As stated by Mate, it should be through extra specs like " hardware_resoruce:NIC_VF_1G=1". We can provide a new filer to make sure the cloud can meet such requirement, or, we can extend current compute capability filter. Also such information will be kept in the instance entry, so that corresponding handler will be invoked when the instance created.

	4) How the cloud user get idea of available hardware resources? An extension will be provided so that novaclient/dashboard can fetch the hardware resource types, resource descriptions, and how much resources left etc.

	Any idea/feedback on this?

Thanks
--jyh