I record here a scheme that was not implemented. It was imagined as a solution for a data center running several instances of Keykos. It is still sometime necessary to take a system out of service, either on a planned or unplanned basis. It is often the case that extra capacity is available on other hardware systems that are running Keykos. This is slick for the case of planned hardware service during times of light load. In case of unplanned outage, load shedding might be arranged subject to service level agreements. It is generally possible to arrange hardware so that the other systems can access the disks of the CPU that is out of service.

The idea is for one computer (= CPUs + RAM) to serve the load of the unavailable system. As the guest disks are introduced to the host system the host system kernel recognizes them as such and instituted a simple mapping of the CDAs. This map is applied in either direction for pages and pots coming and going to disk. This is possible because only the kernel knows where the CDAs are. The Keybits function would also be wary of the map to preserve the illusion of the few guest user mode programs that think that they deal with CDA’s. This might be done conceptually by creating a distinct keybits object for the guest.

There is one big problem when it comes time for the guest to leave: Guest’s pages and nodes may be scattered among those of the host in the host’s swapping area. It might be possible to always swap the guest pages and nodes on the guest’s disks, But my sense of the kernel logic is that it would be easier to flush the swapped guest material to its home position. Some variation on migration logic can cause this. The current kernel is noteworthy for having no such function now.