skas3 versus skas0

skas3 versus skas0

The previous chapter contained a discussion of whether to leave the host unmodified and have the UML instances running in skas0 mode or to patch the host with the skas3 patch for better performance. Since we're now talking about a large UML server and we're trying to get every bit of UML hosting capacity from it, I recommend patching the host with the skas3 patch.

The reasons were mostly covered in the discussion in the last chapter. You'll get better performance with skas3 than skas0 for the following reasons.

• skas3 creates one host process per UML processor while skas0 creates one per UML process. This consumes host kernel memory unnecessarily and slows down process creation.

• skas3 page faulting performance is better because it has a more efficient way to get page fault information from the host and to update the process address space in response to those page faults.

In addition to better performance, skas3 will have somewhat lower host resource consumption due to the smaller number of processes created on the host.