On Thu, Sep 06, 2001 at 03:50:01PM +0200, Robert Waldner wrote: > > On Thu, 06 Sep 2001 09:34:05 EDT, Peter Billson writes: > >> cat /proc/meminfo > >> cat /proc/loadavg > > > > The meminfo would help him but he posted that he didn't understand load > >average and, anyway, needs percent of CPU used. You can not calculate > >CPU usage from load average. > > Not to mention the deep dark magic by which loadavg is generated. I > still don´t understand that completely ;-) > > And yep, that should´ve read > cat /proc/stat > instead. (And no, I don´t know what the values in the first line > exactly mean, but as soon as I set up mrtg again, I´m gonna read up on > the kernel-sources) The problem with using /proc/stat is the values presented on the cpu line are running totals of jiffies spent in user, nice, system, and idle respectively. So, you have to read /proc/stat at least twice and then calculate the deltas (and average over the time delta if you seek an average. Reading man proc and the source code (proc_misc.c; function kstat_read_proc) helped a lot; man proc seems a bit out of date WRT 2.4.x kernels. Yesterday I wrote a perl script that does this (I'm playing with cricket ... see http://canaris.visionary.micromuse.com/cgi-bin/cricket/grapher.cgi?target=%2Fservers I'll make the script source available if someone wants it ... I use a db file to store the readings from each run for use in the next run. My loadavg figures come from /proc/loadavg ... I wasn't interested in any heavy lifting :) Now for the mem stats ... -- Nathan Norman - Staff Engineer | A good plan today is better Micromuse Ltd. | than a perfect plan tomorrow. mailto:nnorman@micromuse.com | -- Patton
Attachment:
pgpTrJPXLx6n6.pgp
Description: PGP signature