Re: Narrowed down bottlneck to disk, how can I find out which FS is being hammered? 100% utilization

From: Tonij (tonij67_at_hotmail.com)
Date: 04/28/04 

Date: 28 Apr 2004 06:17:10 -0700

Darren Dunham <ddunham@redwood.taos.com> wrote in message news:<66wjc.56908$Em6.22947@newssvr25.news.prodigy.com>...
> Tonij <tonij67@hotmail.com> wrote:
> > In my process of drilling down to get to the bottom of our performance
> > issues, I am seeing huge I/O wait on a particular disk. What tools
> > are available to dig deeper, find out which file system is hammering
> > the disk and/or which processes are hogging the CPUS? We are talking
> > 100% utilization here:
>
> "hogging the CPUs"?
>
> > mpstat output:
>> Your CPUs are nearly idle. You're only averaging around 15-20%
> utilization.

They are at 100% all day long, the format of the mpstat got messed up:

mjf xcal intr ithr csw icsw migr smtx srw syscl usr sys wt idl
0 3283 288 1 5986 318 657 87 0 6553 25 23 52 0
0 4283 3895 3791 5532 256 462 244 0 6398 33 17 50 0

Am I not reading that correctly? The last column "idl" is telling me
the cpus are spending 0% of the time idle, or 100% busy.

>
> I'd look at this page:
> http://www.sun.com/sun-on-net/itworld/UIR961001perf.html
>

Will do.

> It's not identical to your situation, but it shows the technique of
> using 'prex' to find processes that are writing to a disk. If you dump
> the output and write short script to compile a histogram of the
> processes that are doing I/O, you might have a hog show up.
>
> > This system looks like this all day long! Main apps are Oracle and a
> > thirdparty app that I have not yet figured out what it's purpose is (I
> > got thrown into a new environment where the previous sysadmin left,
> > you know the story :)
>
> Probably Oracle. Are you a DBA? Do you have anyone that can see if
> Oracle is performing well or not?

I am not a DBA. The DBAS telle me "things run slow" :/
>
> > I have scripts running to capture sar, vmstat, iostat, mpstat, netstat
> > output and they all point to i/o wait...
>
> Great. That's what databases do. They move data around. If you don't
> have anything else to do, then the CPU will appear in I/O wait. That's
> not necessarily a problem.

Not sure I understand this...if my CPUs are at 100% all day long and
disks are waiting it is not a problem?

TIA,

Relevant Pages

  • Re: Narrowed down bottlneck to disk, how can I find out which FS is being hammered? 100% utilization
    ... I am seeing huge I/O wait on a particular disk. ... > the disk and/or which processes are hogging the CPUS? ... processes that are doing I/O, you might have a hog show up. ... Oracle is performing well or not? ...
    (comp.unix.solaris)
  • Re: CPU waiting for... what? (mistery)
    ... > size of the Oracle buffer cache. ... iostat command) gets about tenfold. ... Thank you for your explanation about the steps Oracle uses in this ... > A VERY large fraction of it is single block I/O. ...
    (comp.unix.solaris)
  • Summary: Sizing Hardware
    ... Oracle treats one UIV as actually being two CPUs, ... Will the V1280 cope as a database and web server? ... CPUs are at 1 or 1.2 GHz. ...
    (SunManagers)
  • Re: ra_pages and io rate
    ... -asyn I/O package source) but constantly deliver a high load ... -the cache of the I/O subsystem - on average, ... Two of the oracle systems which i am ... data from the disks, the ...
    (comp.databases.informix)
  • Re: List of running processes in C++
    ...  The CPU usage is usually negligible. ... Which basically means that oracle does the writing for you, ... process's I/O bandwidth is controlled by oracle's I/O bandwidth. ...
    (comp.unix.programmer)
Loading