On Monday, 12/14/2009 at 10:49 EST, Kris Buelens <kris.buel...@gmail.com> wrote: > Two Alan's, one is right, the other... > > There were APs and MPs, Attached Processors, Multi Processors. Ages ago, I > explained both in a course and the difference was that the second processor in > an AP config couldn't perform I/O; on an MP, both could perform I/O, but each > only to the channels it had.
This Alan was wrong and I yield the field. Indeed, I was conflating AP and MP. Yes, in AP, there was only one channel set (16 channels) and it belonged the base processor. (At the time, processor design was that the I/O bus was bound to a single CPU.) MP had one channel set per processor. There were eventually instructions and connectivity added that enabled a "channel set takeover" by the AP in the event of a CPU failure. So, a shared bus, but only one CPU could use it at a time. According to an informant, VM/XA could simulate 370 AP. You could DEFINE CPU 1, but SIO/SIOF didn't work. Since the machine was actually running in S/370-XA, there was no real AP, so it was a case of CP kneecapping the virtual AP to stop SIO(F). Alan Altmark z/VM Development IBM Endicott