Would IBM care to comment to this group concerning CMMA?
I am about to embark on a benchmark of CMM and CMMA.  During my initial playing 
with setup of CMM and CMMA, I have found CMMA to be much easier.
Being lazy I like the Ronco oven approach ( sorry if you don't watch 
infomercials) of "set it and forget it". I too want performance data so don't 
misinterpret the
previous sentence.
 
I like the idea of linux checking/setting extended hardware bits and then DIAG 
10 RELPAGing. CMMA is just CMS on steroids.
With the ballooning approach of CMM there is more work to do, plus I wouldn't 
rush into using production servers on the shared memory pool. 
I want favored nation status for production servers.
 
Something appeals to me about the intrinsic nature of having the kernel decide 
which page to DIAG 10 when CMMA is used.  I realize that there are implications 
of doing this
for System z far beyond our mainframe shores but frankly I don't care.  We need 
all the help we can get managing our expensive memory plus we have to deal with
our naughty WAS children that never drop from queue.  I'm expecting big things 
from CMM and/or CMMA.
 
David Kreuter

________________________________

From: Linux on 390 Port on behalf of Rob van der Heij
Sent: Wed 12/17/2008 11:16 AM
To: [email protected]
Subject: Re: Over Committing Storage in z/VM



2008/12/17 ???? ???? <[email protected]>:

> Anyone here using CMM for over committing? It sounds great! is it working
> great as well?

It's so great there are even two of them ;-)

CMM-1 (also called ballooning) is to temporarily take away memory from
Linux when it should not be using it. Can be used if you understand
the application requirements in Linux, for example with a TSM backup
that should have memory over night but not during daytime. Or when you
have an external process that can take VM performance data and decide
how much Linux can have or needs.

CMM-2 (or CMMA -  requires adequate hardware and software)  This is
supposed to happen all automatic with no tuning options and no
instrumentation on what it does... Current software levels don't
enable it per default, and future levels may not include the full
feature anymore.
For CMMA, I used to say "It is a very cool research item. In lab
environment, an artificial workload could be constructed that takes
advantage of it."  Any performance data I have seen from IBM confirms
this status.

And for completeness, there is CMM-0 ;-)  Make sure your virtual
machines drop from queue when idle, so z/VM can do memory management.
Tune the ratio between high-performance swap and virtual machine
memory based on the workload characteristics and business
justification (so a test server has higher latency and lower cost).
Look for my post on Nov 27th:
www.mail-archive.com/[email protected]/msg50668.html
(Google found it searching for ""real disks for swap is only good for
slowing down Linux")

Rob

Rob van der Heij
Velocity Software
http://www.velocitysoftware.com/



----------------------------------------------------------------------
For LINUX-390 subscribe / signoff / archive access instructions,
send email to [email protected] with the message: INFO LINUX-390 or visit
http://www.marist.edu/htbin/wlvindex?LINUX-390

Reply via email to