On Thu, Dec 15, 2011 at 4:46 PM, David Boyes <[email protected]> wrote:
>> Quick question.  When you run top on RHEL5 linux on systemz, is it accurate
>> or is it a best guess?
>
> It's mostly not very useful. It reflects only the usage when the virtual 
> machine is dispatched. I/O counts are "mostly" OK in that it's a reflection 
> of virtual I/O requests (does not reflect that CP may have done something 
> smart with the I/O to make it more efficient -- think about VDISKs for a 
> moment). Memory usage is from the perspective of inside the virtual machine 
> (doesn't reflect impact on the total machine).

You dismiss CPU usage too quick as "mostly" OK - while vague, probably
still not useful.

* Recent kernels the Linux CPU measurements only show "virtual" time,
it misses the overhead. Thanks to virtualization hardware, you get
close. But in this configuration with Linux running on second level
z/VM, the overhead is significant (read: high). So "top" inside that
Linux will not show you the amount of CPU resources consumed to run
the guest.

* Most people using "top" don't really care about what they use, but
about what they don't use. The assumption is that what you don't use
is still available to use. But not in a virtualized environment...

You can't have it all, and in this case you can't have either of them.
For the full picture you need a performance monitor that combines data
from inside and outside the virtual machine.
-- 
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
----------------------------------------------------------------------
For more information on Linux on System z, visit
http://wiki.linuxvm.org/

Reply via email to