Ultra1 Crash - FIX?

[Robert Dinse <nanook@eskimo.com>]

     Ultra1 crashed again this evening.

     Ultra1 is an Ultra 1 with 640MB of RAM and two 30GB IBM SCSI disks
presently running Linux 2.2.17 with RAID and NFS patches. 

     I thought it would be good to put web server on the machine with the files
since it eliminates NFS as a weak link, but for some reason the combination of
Apache/Sparc/Linux seems to be unstable.

     This machine ran for months at a time before Apache was put on it even
with the above mentioned RAID and NFS patches. 

     I'm going to move the web server to a different platform because I've
become convinced the Linux/Sparc/Apache combination can not be made to be
stable.  We had stability problems on 32-bit Sparc, Dave Miller said, "Go to
Ultra you won't have these problems", I did, problems persist although at a
slightly attenuated level. 

     Failure modes include running out of memory even though there are 640MB,
and this in part is related to bandwidth (and we have another T1 on order that
should rectify that), but other failure modes include "Failure to dereference
NULL pointer" which was also a frequent crash on the 32-bit hardware and a
state where the machine is still alive in terms of echoing keystrokes and you
can switch virtual consoles, but does not respond to anything else. 

     I'd be very interested to hear from people running Apache web servers that
are doing in the neighborhood of 2 million hits/day or better and are stable.
By stable I mean averaging at least one month between crashes with a load of
two million hits per day or better.