[NCLUG] suggestions on building a box

[Sean Reifschneider]

On Fri, Jan 12, 2001 at 03:29:59PM -0700, Matt Rosing wrote:
>The advantage to the burn-in is to limit the chance of DOA.

Uh, if the component is DoA, you probably don't want to run a burn-in on
it.  You're safe just sending it back and asking for a new one.  Chances
of a DoA component suddenly coming back alive on you are ... slim.  ;-)

The reason for doing a burnin is to ensure that all the components of the
system work together when the system is under various kinds of load.  My
burnin test is simple.  I pick up the SRPM for gcc, and simply set up a
script that loops doing "time rpm --rebuild gcc*.src.rpm" and pipe this
out to a file.  After it's run a day or two, I look at the log file and see
how many "laps" it's done, and what the times are on each lap.

If any lap is significantly smaller than another, there was probably a
problem...  Like if it takes 22 minutes normally, and one time out of
5 it takes less than 5 minutes, you know it's dieing.  I also look in
the log for occurances of "signal", but with gcc that means you pick up
things like "signal.c" too.  You need to grep those out.

This catches the "signal 11" issues and the like.  It's proven very
reliable for me, particularly when I was mucking around with hand-drilled
Celerons modded to SMP running 300As at 450MHz...

Sean
-- 
 ISA isn't dead, it's just that people wish it were.  The correct term for
 this condition is "legacy"...  --  Sean Reifschneider, 1999
Sean Reifschneider, Inimitably Superfluous <jafo at tummy.com>
tummy.com - Linux Consulting since 1995. Qmail, KRUD, Firewalls, Python