Intel Rack Scale Design (RSD) has some concepts like "Disaggregated Hardware" and allows "Node Composition", but I am not able to find what exactly this is or how it works.
EG If a rack has 4 servers, each with its own Processors and Memory and Disks and Networking, then:
Can we compose a node with some subset of hardware from a single server and make a composed (virtual?) node ? What OS will the composed node have ?
Can we compose a (virtual?) node with hardware from multiple servers ? If yes, how do the multiple servers share the hardware ?
When we make the virtual nodes, is it using some new virtualization technology by Intel ? Or is it using existing technologies like Oracle VirtualBox or VMWare ESX ?
How RSD works when the 4 physical servers in the rack have a mix of Windows & Ubuntu & CentOS installed ? Which OSes are compatible with RSD ?
(( Newbie alert : this question may sound ambiguous because I do not know what exact question to ask because RSD Documentation is sparse ; I am open to suggestions from experts ))
In my opinion, "Node Composing" in intelRSD is not real virtualization. As mentioned in "pod-manager-api-specification":
Because the RSD is designed to access HW resource only via remote methods(IPMI, iPxe deep discovery), and virtualization can not be conducted on bare metal, I think the RSD Composing node just simply marks the HW resource as "Allocated" and specifies an iSCSI target to access remote drives.
Maybe you can create an issue on the official repository, I am also looking forward to the official explanation. https://github.com/01org/intelRSD