linux vmstat "steal" column

[Dean, David (I/S)]

Is there a way to see what CP is doing during "steals"?

David M. Dean
Information Systems
BlueCross BlueShield Tennnessee




-
Please see the following link for the BlueCross BlueShield of Tennessee E-mail 
disclaimer:  http://www.bcbst.com/email_disclaimer.shtm

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

Re: linux vmstat "steal" column

[Shane G]

Hopefully doing its job - servicing other guests. What are you asking ?.
Linux metrics are from the perspective of Linux - and it has a heritage of a 
stand-alone O/S. It may 
think the cycles have been "stolen", others may disagree.
Think of it as involuntary non-dispatch. Has to happen in any non bare-metal 
environment - including 
(non-z/VM) PR/SM.
Things are going to get even more confusing if the distro distributors decide 
to include the "guest" 
numbers that are maintained on current kernels. Especially if Redhat (for 
example) decide to ship a (z 
series) KVM option...

Shane ...

On Fri, May 7th, 2010 at 4:02 AM, "Dean, David (I/S)"  
wrote:

> Is there a way to see what CP is doing during "steals"?

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

Re: linux vmstat "steal" column

[Alan Ackerman]

Why do you want to know what CP is doing? There really isn't any way that I 
know of to find out. You could turn on various traces, and then write code to 
post-process those traces and see what CP is doing. (Or ask IBM to process 
them.) But unfortunately, turning on the traces would probably slow things down 
enough that the results would be quite different from the situation you started 
with. And in the end. what CP is doing might not be very interesting or useful. 
It would certainly require a fair knowledge of CP internals, as well.

A much more useful thing is to ask is what your particular Linux guest is 
waiting for. z/VM has a "high speed state sampling" component that provides 
this information in the monitor records. I can read this in the ESAMON ESAXACT 
panel. (I leave as an exercise for it to someone else to translate ESAMON to 
other monitors.) You can certainly read CPU wait there. If CPU wait is higher 
than you like for your Linux guest, you can use SHARE settings to make it run 
faster -- but this is  a case of "robbing Peter to pay Paul". To lower CPU wait 
for one guest means raising it for other guests. Or you could add additional 
engines to your z/VM system.

See  for a sample ESAXACT panel.

There are other wait states in the ESAXACT panel -- I/O wait, page wait, etc. 
Some of them that are not as easily understood. If you really have a problem 
with what you see, you may have to contact IBM or your monitor vendor. (I don't 
know any way to get this information without a monitor. You could write your 
own code to post-process monitor records, I suppose. You would learn a lot, but 
it would be rather time-consuming.)

It would be interesting to see how the numbers that Linux reports compare to 
the ones that CP reports -- and to try to explain the differences. I certainly 
don't have time for the exercise, but maybe someone else has done it? I would 
believe the z/VM numbers over the Linux numbers, simply because z/VM knows more 
about what is actually happening than Linux does.


On May 6, 2010, at 11:02 AM, Dean, David (I/S) wrote:

> Is there a way to see what CP is doing during "steals"?
> 
> David M. Dean
> Information Systems
> BlueCross BlueShield Tennnessee
> 
> 
> 
> 
> -
> Please see the following link for the BlueCross BlueShield of Tennessee 
> E-mail disclaimer:  http://www.bcbst.com/email_disclaimer.shtm
> 
> --
> For LINUX-390 subscribe / signoff / archive access instructions,
> send email to lists...@vm.marist.edu with the message: INFO LINUX-390 or visit
> http://www.marist.edu/htbin/wlvindex?LINUX-390

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