> Date: Tue, 11 Jun 1996 01:48:42 +0200 (MET DST)
> From: Mikael Abrahamsson <mike@uplift.sparta.lu.se>
> To: Aaron Tiensivu <tiensivu@pilot.msu.edu>
> Cc: Jeff Johnson <trn@gate.net>, linux-kernel@vger.rutgers.edu
> Subject: Re: 2.0, loggings, cpu quotas, 2.1 issues, etc.
>
> uOn Mon, 10 Jun 1996, Aaron Tiensivu wrote:
>
> > > ***** Second, I would like to see some of the meter-0.2 or Portato code
> > > integrated into kernel source and/or modules. This way, system status
> > > lights could be updated by the kernel, without reading /proc or calling
> > > uptime/ps/the like. This might be considered bloat, but that why it is
> > > optional.
> >
> > I think this still belongs in "user-space-land".. that's a featurism that
> > probably .5% of the Linux community uses and works well, in user-space.
> >
> > That's the main reason /proc is there, from what I can gather.. so you can
> > write 'support' programs that can do exactly what 'Portato' or 'ps' does.
>
> As of 1.3.73 (or around there) and up, running portato gives a very high
> load in my opinion. Bloats my "xload" :) I stopped running it a few days
> ago when I discovered that. I like the util otherwize, and if it could be
> implemented somehow so that it doesnt use as much CPU time, I'd appreciate
> it. I'd say the load rises by 0.2-0.5 when starting portato. When you "ps
> -ux" it, or use "top", it doesnt show as using a lot of CPU, so I dont
> know why the load averages go up as much.
>
> Just seems to me it would benefit from being put into the interrupt
> handler, and reading the load averages from within the kernel structures.
> Would probably be just a few lines of code run every jiffy.
The reason that the load is upped so much is because it is opening and
closing files in /proc all of the time, sometimes even twice a second,
giving a low CPU usage in user time, but lots of system time as the kernel
performs the actions.
If this was included in the kernel, it would take much less CPU and work
better. Sort of like the watchdog timer, even though I still use loaddog,
it works better in the kernel.
Oh, that is what the syslog problem was... loaddog was running 'date' to
check for new processes. Doh! :-)
And no one is talking about my CPU quotas yet.