There is a ton (metric or otherwise) of information in the monitor validating effects of
CMM. But if you don't have ESALPS, I'm clueless about what you would look at. On ESALPS,
there's user perspective, there's page space perspective, there's page rates, there's page
movement above/below the lines.
Could I recommend taking a class in z/VM and Linux performance? as well as a performance
monitor?
"http://velocitysoftware.com/workshop.html";
Brian Nielsen wrote:
On Wed, 12 Dec 2007 09:58:04 +0930, Fred Schmidt <[EMAIL PROTECTED]>
wrote:
However, I would like to persist with CMM. My original question still
stands... how can I verify that it is actually working?
Well, there are 2 answers to this.
First, the Linux guest has issued at least one ESSA instruction as
evidenced by Q MEMASSIST reporting "ACTIVE". The sole purpose of ESSA is
to implement CMMA. Since the virtual machine didn't get an operation
exception on the ESSA, it was either handled by the hardware assist or
simulated by CP. Personally, I'm reasonably sure that either did what it
was supposed to do with the ESSA, but Your Milage May Vary. I don't know
if there are any stats in CP or Linux on pages freed via CMMA.
Second, you could compare performance stats with it turned off and turned
on. With it on you would expect CP to do less paging and Linux to use
more CPU. See the IBM white paper "z/VM Large Memory – Linux on System
z"
for more detailed info:
http://www-03.ibm.com/support/techdocs/atsmastr.nsf/WebIndex/WP101151
Brian Nielsen
