Well, what a strange tale to tell.... For your Friday enjoyment. First
off, let me thank all the responders:
Rahul Roy <roy@bluestone.COM>
bismark@alta.Jpl.Nasa.Gov (Bismark Espinoza)
Jim Harmon <jharmon@telecnnct.com>
Alan Chan <A.Chan@CdnAir.CA>
djohnson@nbserv2.dseg.ti.com (Danny Johnson)
Kevin.Sheehan@uniq.com.au (Kevin Sheehan {Consulting Poster Child})
Dan Pritts <danno@aa.fv.com>
Lee Hughes <lhughes@anitesystems.co.uk>
Gnuchev Fedor <qwe@ht.eimb.rssi.ru>
gibian@stars1.hanscom.af.mil (Marc S. Gibian)
"Andrew Moffat" <amof@SubaruSparcDev.subaru1.com>
bill@aloft.micro.lucent.com (Bill Shorter - Local Account)
negativl@netcom.com (Raymond Wong)
jeffw@smoe.org (Jeff Wasilko)
Jochen Bern <bern@penthesilea.uni-trier.de>
And here's the original request/problem and followup:
-----------------------original---------------------------------------------
I can hear it, perfmeter SEES it, so where is it coming from???? Perfmeter
shows disk io at about 200 (io/sec?) and I can hear it clacking away but the
system is really pretty idle. vmstat doesn't show much disk activity at all,
mostly zero but it DOES show lots of 'in' and 'sy' (40-50 and 200-400
respectively). I've done some traces on all the processes that seem to
be visible etc. with no luck.
This is SunOS 4.1.3. on a SS2. Any ideas how I can find out where the disk
io is coming from???
Tnx, I'll summarize.
Bob
----------------------------------------------------------------------------
Some ideas have come in and indicating that I should have given more info:
perfmeter shows ONLY disk traffic as being high. swap and paging are
very low 0-3 per interval so it would seem that is not the problem.
iostat (iostat -d 5 or -D 5) shows very little activity at all for
the two disks (!!??)
memory - with a few light users (pine sitting idle for 3 or 4 users
for example) there is about 9 megs available memory (of 64)
pstat -s shows 24320k alloc + 5904k resvd = 30224 kin use and
51092k avail. (total swap is 81 meg)
Someone suggested lsof but that just shows files open and who is using them.
What I could use would be some sort of IO count per file or per process or
something....
ps shows only pagedaemon and swapper in "D" state which seems to be sorta
normal (?)
Some suggested NFS. I'm gonna look into that but I only mount a couple
of file systems from another machine and export none from this one...
--------------------------End of Original-----------------------------------
A number of people answered with various ideas about vmstat, iostat, fuser,
the possibility it was swap space size problems, NFS traffic etc. etc.
All of which might have helped >IF< the problem had been what I originally
claimed - disk noise/activity and perfmeter showing it. Turns out that
NEITHER was the case.
As to the noise, "Andrew Moffat" <amof@SubaruSparcDev.subaru1.com>
questioned that premise:
---------included message----------------
Hmmmm. Call me old fashioned, but I'd trust iostat more than perfmeter. To
me, there appears to be more evidence indicating that the system really is
quiet, than indicating otherwise. I'd start doubting perfmeter and the noise
you're hearing. Are you sure its not a noisy fan or something you've got there?
---------end included message-------------
Well, son-of-a-gun... Due to the way my workspace is laid out the
acoustics are such that it SOUNDED like it was coming from the disk but in
reality it was the fan on the 8mm tape drive making the noise.... (!) What
I did was go to the disk box (on a shelf under the table top) and opened the
side panel to see if there was maybe an activity light on the disk. With
my head down there I couldn't hear any activity... He was right - it was
a fan, not the disk.
But I STILL had the issue of why perfmeter is all those disk IOs when
iostat and company are showing idle.
jeffw@smoe.org (Jeff Wasilko) gave me the first clue:
---------included message----------------
I seem to remember that for some reason Solaris 1.x puts serial
I/O into the disk category of the perfmeter....Perhaps that's
it?
---------end included message-------------
That was close, as it turned out, it IS something other than disk,
but since I have nothing going on the serial ports that wasn't it.
Then, Jochen Bern <bern@penthesilea.uni-trier.de> hit it on the head:
---------included message----------------
Wild Idea:
I once read about perfmeter erroneously taking Keyboard Interrupts
(including Mouse Movement) as Part of some other Type of Activity.
I don't remember whether it was "disk" Activity - but I just tried,
when I start moving my Mouse around like crazy my "disk" Bar Chart
goes from nearly nothing to 320 (SS10 running 4.1.3_U1B).
---------end included message-------------
That was it! For whatever reason, mouse activity gets counted as disk
traffic.
Thanks again to all who responded.
Bob
-- ---------------------------------------------------------------------------- |Bob Rahe, Delaware Tech&Comm Coll. / | |Computer Center, Dover, Delaware / | |Internet: bob@hobbes.dtcc.edu / | ----------------------------------------------------------------------------
This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:11:56 CDT