Re: /proc/loadavg disagrees with top and ps
Paul Gear <paul@gear.dyndns.org> said on Wed, 04 Aug 2004 07:03:12 +1000:
> This is an OpenPGP/MIME signed message (RFC 2440 and 3156)
> --------------enig162A5A009C607900848B2DE4
> Content-Type: text/plain; charset=us-ascii
> Content-Transfer-Encoding: 7bit
>
> Reid Priedhorsky wrote:
> > Hello everyone,
> >
> > /proc/loadavg currently reports the following:
> >
> > 0.96 0.98 0.78 1/116 23994
> >
> > xload also reports roughly the same.
> >
> > But top and ps both report a nearly idle system (98% idle). What is going
> > on? How can I find out what is causing my system to be so busy?
>
> I've seen load averages of 14 and 16 when the CPU usage was on 10%. The
> two are usually, but not necessarily, related. There are certain types
> of work where this behaviour will be seen. Even low amounts of I/O to a
> slow device, if done by enough processes could cause this.
As a test one day, I mounted nfs over the modem, and ran about 300
processes doing a find over the modem. CPU usage was ~10%, 15 minute
load was above 200 :)
As opposed to undergrad, where in the last 2 days before the semester
project was due, when the lusers discovered they might need to think
about *starting* their projects, the load on the 4 poor 8 proc sun
boxen would hit 373. We actually witnessed a wraparound at some point,
where the load seemed to go down to 8, yet intereactivity was still
thouroughly poor.
At that time, I tried to help someone track down a missing brace in
their C code, so I fired up emacs, waited 8 minutes, pressed C-x h
C-M-\ and waited for another half hour before giving up and leaving
her to fend for herself :)
Oh - and the waiting 5 seconds for your bash *shell* to echo a single
character keypress. <shudder>.
--
TimC -- http://astronomy.swin.edu.au/staff/tconnors/
"Does bacteria culture in coffee cup qualify as pet? Have already
givink it name." -- Pitr Dubovich/User Friendly
Reply to: