On Thu, 27 Jan 2005 09:35:59 +0900, Tomoyuki Yatsunami
<[EMAIL PROTECTED]> wrote:
  
> I thought so, too.
> However somewhat this value is around 1ã2 on the client's system.
> I read that when you submit a "vmstat" command, then [cpu us] value
> sometimes could contain other Linux Guest OS's operation, so vmstat is not
> really helpful for performance monitoring. I just wander and want to know
> could this be true to "uptime" too?

I think you are confusing a few things. When Linux reports that it is
idle and is not using CPU cycles (e.g. in sar or vmstat) then that is
true. But since you are monitoring the Linux server with tools running
on the Linux server, there is a small amount of work and you may spot
a process in queue now and then. But the average number of processes
running will be close to 0.

When you have multiple Linux guests on z/VM that all have work to do
and compete for the few real CPU's that you have installed, then that
is going to confuse the meters on Linux. Basically Linux will notice
that it took a minute (wall clock time) to get the work done, so it
assumes that it has been using the CPU for that minute, and reports
accordingly. In reality z/VM has decided during that minute to run
other virtual machines instead of yours. The Linux reports could be
orders of magnitude wrong when reporting how much CPU cycles was used
for some task.

>From what I understand uptime reports the average number of processes
inside this single Linux guest competing for CPU resources. Some of
these processes come from interaction with end-users or requests via
the network. If there is heavy competition for the CPU on z/VM level
then these processes in Linux will take longer to complete, so you
would see more processes queue up inside Linux as well (provided at
least some work does get done in Linux to start those processes in the
first place).

You can even observe this during startup where Linux runs a short loop
to measure the CPU speed. If you boot Linux on a busy day you may get
much lower Bogomips rating than when you boot Linux on an empty z/VM
system.

Now z/VM does knows the real numbers and knows how much CPU resources
the Linux virtual machine really got when it thought it was using a
full engine. ESALPS takes the detailed per-process measurements from
Linux and corrects that with the meters from z/VM. This gets you much
closer to true numbers that can be compared among systems.

Rob
-- 
Rob van der Heij                  rvdheij @ gmail.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