Bill Bitner <[EMAIL PROTECTED]> writes:
> Yes. The one thing to remember there is INDICATE USER gives a
> snapshot of running counters. In most cases, I would use two
> invocations of the INDICATE USER and take deltas.

common folklore is vtime is what the guest would do on the bare
machine w/o vm. however, there have been some exceptions ...  original
vs/1 (and earlier mvt) handshaking basically offloaded some number of
functions out of the guest operating system to VM ... not just because
it eliminated processing duplication ...  but in some cases, the vm
kernel was actually significantly more efficient at performing the
function than what was implemented by the guest operating system.

later there was vm kernel operation restructuring for tpf & 3081. at
the time tpf (airline control program) didn't have shared-memory
multiprocessor support (and 3083 hadn't been retrofitted to 308x line,
originally 308x was never to have a uniprocessor offering). nominally
the vm kernel overhead ran serially with guest operation ... however
running 3081 multiprocessor with a single tpf guest ... resulted in
one processor being idle. the vm kernel was restructured to introduce
a queueing and signalling mechanism for some number of operations that
vm must do on behalf of the virtual machine. in some number of cases,
the increased the absolute processing cycles (because of the
addditional queueing and signalling) ... but if it turned out to be
the single tpf guest case on a 3081 ... some amount of the vm kernel
processing could now go on in parallel and asyncronously on the 2nd
processor and allowed the vm kernel to more quickly return control to
the guest operating system.

if the customer was already running all processors at max ... the
restructuring done in that release ... degraded overall thruput ...
but for the single tpf guest case (not having smp support) on a
two-processor 3081, it allowed for overlapped processing on the
otherwise idle 2nd processor.

--
Anne & Lynn Wheeler | http://www.garlic.com/~lynn/

Reply via email to