Re: run in queue (Solaris)

Hi and thanks for the excelent answer.

What i´ve did yesterday was simply a /usr/ucb/ps -aux to see every active process and in that way somehow search för the process that was taking up all resources, and after finding this complex query that seemed to run forever i´ve simply killed it (kill -9), maybe i should have used the "alter system kill session" command instead..

But the system still isn´t very healthy since we in the daytime have about 15 processes in the run queue on our 2 processes database server and this day i´ve wasen´t really able to connect any process to the high load on "run in queue"...

Is this the best way to find perfomance bottlenecks..?

Someone that would like to share their "perfomance fault isolation method" with me..?
Regards
Jens

"koert54" <koert54_at_nospam.com> wrote in message news:Pqup7.59215$6x5.12755875_at_afrodite.telenet-ops.be...
> Forgot one thing - if there was only one query running when you were
seeing
> a runqueue of 30, most probably the query
> used PQ slaves ... I've seen systems with large degree of parallel degree
> generating a large runqueue ... now, if you 'simply' killed
> the query what exactly do you mean... a kill -9 of the dedicated server or
> an alter system kill session. If you kill the query coordinator
> with a kill -9 I don't know how Oracle would react but I guess SMON or
PMON
> should detect this and clean up your mess.
> This could explain why your runqueue does not drastically decrease
immediate
> after your kill but only after a while...
> BTW - the command uptime actually displays the average runqueue length
over
> 5min, 10min, 15min period ...
>
> Cheers,
> Koert
>
> "koert54" <koert54_at_nospam.com> wrote in message
> news:9jup7.59207$6x5.12753057_at_afrodite.telenet-ops.be...
> > The r field in vmstat is the count of processes that are currently
> runnable.
> > However, depending on the platform specific implementation, this may or
> may
> > not include
> > the number of jobs running on the CPU(s).Example - SunOS 4.1(and most
BSD
> > implementations)
> > include both the number of jobs running on the CPUs and those waiting in
> the
> > queue. Solaris on the
> > other hand excludes the running jobs.
> > Say you have 5 runnable jobs on a 2way box :
> > - running SunOS 4.1 : the r column shows 5
> > - running Solaris : the r column shows 3
> >
> > The r column is a very good indicator if the machine is CPU bound. If
you
> > constantly having 5 runnable jobs
> > on a 2 way box, adding 3 CPU will prove very valiable and should boost
> > performance. However - having
> > constantly 2 runnable jobs on a 12way box is a waste of money ... 2
CPU's
> > would have been sufficient.
> > Typically, a runqueue of 3 to 4 times the number of CPU's will bring
your
> > machine to a crawl because the CPUs
> > are forced to jump between the runnable jobs.
> > In your case a runqueu of 30 is a lot if you only have 4 CPU's - but ok
if
> > you 24CPUs...
> >
> > A great book on this matter (besides 'The design of the Unix operating
> > system') is "Optimizing Unix for Performance"
> > by Amir H. Majidimehr.
> >
>

http://www.amazon.com/exec/obidos/ASIN/0131115510/qid=1000763299/sr=2-2/ref=
> > aps_sr_b_1_2/002-2667980-3793602
> >
> > Cheers
> > Koert
> >
> > "Jens" <superglad7_at_hotmail.com> wrote in message
> > news:TWtp7.5603$sn6.611353_at_newsc.telia.net...
> > > Hi all
> > >
> > > This question should maybe go to a Solaris newsgroup but i´ve hope
some
> > > DBA´s out there could answer this one pretty quick and easy..
> > > To the question:
> > >
> > > When running a "vmstat" on my Solaris 8 i could see a lot of "run in
> > queue",
> > > about 30 or
> > > more!
> > >
> > > I´ve searched for processes and found one process containing a suspect
> > query
> > > to the Oracle database wich contained a lot of wildcards and so on
wich
> > i´ve
> > > simply killed.
> > > Afer this the system answered much faster.
> > >
> > > But, now to my question, what exact does "run in queue" tell me..?
> > > After killing the process the "run in queue" wasen´t down to zero
> > > immediately but slow decreasing.
> > > Now, after alomst 30 minutes the run in queue is down to about 3 or so
> and
> > > soon to be zero if i´ve know the system right..
> > > Is this what i´ve should expect or more exactly, coul´d someone please
> > tell
> > > me a bit more about "run in queue" and what this really means..?
> > > Regards
> > > Jens
> > >
> > >
> > >
> > >
> >
> >
>
>
Received on Tue Sep 18 2001 - 13:18:52 CDT