Alan, ESALPS correlates the linux process data and the z/vm data,
allowing chargeback to be done correctly at the process level. Other
products have not announced this capability as far as i know? So you
would be correct for other methods of collecting process data.
Alan Ackerman wrote:
On Thu, 2 Apr 2009 09:15:57 -0500, =?iso-8859-1?Q?Greg_Dyrda?=
<gregory.l.dy...@us.hsbc.com> wrote:
We currently bill for Linux on a per guest basis. I'm wondering what
approach others are taking. Specifically, I'm wondering if it is
possible
to bill at the process level and if anyone else is billing that way.
========================
=========================
=======================
I think a Linux expert would have to answer this, but I don't think it is
possible to charge-back accurately on a per-process basis when running
under VM. The actual CPU time is captured by VM (CP) -- but it knows
nothing about processes. The CPU utilization numbers per process that
Linux produces in, e.g., TOP, are just plain wrong, because Linux does no
t
notice when the processor has been taken away and given to another guests
.
Linux thinks elapsed time while process is running = CPU time -- but of
course it does not.
I do remember information at SHARE about revisions to Linux to allow it t
o
know what the real processor utilization is (by issuing the appropriate
DIAG instruction), but I don't know if that ever made it into the Linux
shipped by Red Hat or SuSE/Novell.
That’s why I said a Linux under VM expert would need to answer this.
Right now we are working on moving VM accounting data to MICS on z/OS to
do charge-backs, and that is by virtual machine, not by process.
You could use the per-process CPU times in Linux to prorate the VM
measured CPU time, but I doubt that would be accurate. Some processes are
much more likely to have the CPU stolen than others. (Those that are CPU-
bound instead of I/O-bound, for example.)
Alan Ackerman
Alan (dot) Ackerman (at) Bank of America (dot) com