[Orca-users] Re: Not reporting I/O wait under 2.6?

[Nick Steel (Lists)]

> Hi Blair -
>
> > Nick,
> >
> > I haven't seen this problem myself.
> >
> > Which version of SE are you using on the 2.6 hosts?  Can you try the
> > other example SE scripts and see if they generate the same types of
> > data?
>
> I'm using setoolkit 3.2.1 on all the hosts.
>
> I had a dig through the examples and contribs and didn't find any scripts
> that explicitly show wio.What I did was modify cpustats.se a bit.
> cpustats.se calculated idle by adding wait_time + idle_time together. I
> added a couple more cols to its output to list wait_time and idle_time
> individually. I've had a look at the output on both a Sol8 and Sol6 and it
> looks like its reporting it properly in that its reporting wait_time
higher
> than what the orcallator script is showing.
>
> I had a poke through the percol files and verified that orcallator.se is
> writing out the wrong data.
>
> One thing that's intersting (and I'm not quite sure what to make of this
but
> I thought itw as worth mentioning. I modified orcallator.se 1.28 on one of
> the Sol8 boxes to only look at cpu stats and gave it a very short refresh
> rate and ran it from commant line and the numbers it put out looked right.
> When I tried the same thing on a couple of the 2.6 boxes, I did not get
any
> output at all. I'm begining to wonder if se could be broken on the 2.6
> boxes? Although  the rest of the se scripts in examples/ run without any
> problems.
>
> A very confused
> - Nick
>
> >
> > Orcallator.se uses the SE supplied classes to get this information, so
I'm
> > guessing right now its an issue with SE.

Ok, I'm replying to my own post. I had a poke around with orcallator.se and
I found something quirky with vmglobal_total(); Its setup so that it will
average the .wait_times for all the cpu's toegether if the os minor version
is < 70, otherwise it uses the sum. Here's the code from live_rules.se:

  if (GLOBAL_pvm_ncpus > 1) {
    /* average over cpu count */
    pvm.user_time        /= GLOBAL_pvm_ncpus;
    pvm.system_time      /= GLOBAL_pvm_ncpus;
    pvm.wait_time        /= GLOBAL_pvm_ncpus;
    pvm.idle_time        /= GLOBAL_pvm_ncpus;
#if MINOR_VERSION < 70
    /* reduce wait time - only one CPU can ever be waiting - others are idle
*/
    /* system counts all idle CPUs as waiting if any I/O is outstanding */
    pvm.wait_time        /= GLOBAL_pvm_ncpus;
#endif

I found that if I stopped it from averageing the wait_io the numbers I get
back are much more reasonable and run pretty close to what I'm seeing from
iostat being run at the same time.

Does anyone know if later kernel revisions of 2.6 changed the way wio is
stored and reported that could be affecting vmglobal_total()?

- Nick







> >
> > Best,
> > Blair