[Asterisk-Users] * box hangs after a couple of days...

[Steven Critchfield]

On Tue, 2004-10-12 at 20:54 +0200, Florian Overkamp wrote:
> Hi Steven, 
> 
> > -----Original Message-----
> > You seem to have too much confidence in your raid setup. I'm 
> > wondering if you are using a highpoint semi software raid 
> > since you have mentioned an AMD machine and IDE drives. You 
> > probably are having IDE problems arising from the number of 
> > interupts needed for zap hardware. Looks like you need to go 
> > looking for some replacement hardware.
> 
> This is a rather bold statement. Can you comment on why IDE raid might get a
> conflict in handling irq's with zaptel hardware ?
> 
> We are all aware of the amount of interrupts needed by zaptel, but I've
> handled quite a few boxes with mirrored IDE configs with zaptel cards in
> them, and I've never seen that to be a problem - unless the raid-software
> starts syncing. In that case the zaptel drivers might become a little
> crappy, but I've never seen it crash, even at high loads...
> 
> I would _love_ to create an inventory of issues that can, under reproducable
> conditions, kill asterisk and/or the box it runs on. This might allow all of
> us to learn from each other and see what kind of configs to avoid.

I basically am not impressed with any IDE software raid setup especially
that uses onboard IDE controllers. If the IDE drivers are sub optimal,
the machine could be very stable under most situations. It wouldn't be
fast, but it would be stable. If you start stacking up IDE requests and
they are being regularly interupted by the zap hardware, it might cause
trouble.

Part of the regular suggestions is to remove any hardware that can lock
up the system for an significant time. The frame buffer is a regular
culprit for taking up too much time. Knowing how much performance gain
you can get on a system with mild disk activity when you start working
with hdparm, you have to wonder what kind of hit you get when you have 2
or more waiting for service. 

-- 
Steven Critchfield <critch at basesys.com>