Load average problems

[Dr A V Le Blanc]

I am using arla 0.53.3 on an i386 Linux system with a 2.2.19 kernel.
This is part of a 'farm' of boxes providing a number of virtual
web sites, using apache, with the web pages in AFS.  Anyway,
while it works quite well most of the time, I do occasionally see
the load average on one server box rise to ridiculous levels,
while on the other identical boxes, receiving an equal share of
the incoming connections, this does not happen.  That is, it
seems to affect one or another box for 10 or 15 minutes at a
time, rarely affects more than one box at once, but does occur
at some time or other on all boxes.

I should say this is using glibc (6) version 2.1.3, compiled
with gcc 2.95.2, and the rest of the distribution is fairly
ordinary Debian potato, except that all file systems are reiserfs.

Has anyone else seen unusual averages with arla?  I see arlad
taking up 75% or more of the cpu.  Amon doesn't reveal much;
occasionally workers goes up from 1 to 2 for a moment; the
display above the workers line seems to have nothing to do with
this number.  No other process seems to be using a lot of cpu,
at least not directly; indeed the CPU is fairly idle most of
the time, which strikes me as odd with a high load average.

     -- Owen
     LeBlanc at mcc.ac.uk