[Asterisk-Users] zaptel PRI drivers

[Steven Critchfield]

I am not flaming you even if it may seem so below.

On Sun, 2005-03-20 at 23:12 -0700, Tom wrote:
> I have a quick question.
> I know that running X on an asterisk server is not officially "supported",
> however, I've never had any trouble with it until now (8 months, using wctdm
> cards with fxo and fxs ports, IAX trunks, SIP phones, everything except a PRI
> card).  Now I just installed my first asterisk box that terminates a PRI, and
> bam, HDLC errors up the wazoo if X is running, if its not, everything is fine,
> I assume this is because the timing parameters for the PRI are so much more
> strick.

IT isn't just not "supported", as you see, it just doesn't work
reliably. Stop doing that.

> I don't mind if X is a little less responsive (even alot less responsive), but I
> would really like to be able to run X on a server with a PRI.  Is there any way
> to reduce X11's priority so that it doesn't interfere with the zaptel driver
> for the PRI... I've tried renicing X as far down as I can and renicing Asterisk
> up as far as I can, however I fear this won't ever fix the problem since I
> think the actual kernel module that is running the pri card needs to get higher
> priority (ie, the kernel itself needs higher priority).  Is there any way to do
> this?  Am I correct in my analysis?  I really don't understand why on a system
> that averages less than 3% CPU usage with X running, why it can't handle the
> PRI.  I know for whatever reason X always gets a really high priority (although
> top doesn't show X getting any special treatment its PR 15 NICE 0 by default,
> lower than most other processes on the system).

Your problems our beyond the reach of nice. Bad graphics chips use lots
of CPU time when doing updates and therefore will cause the zaptel
drivers to miss interupts. Missed interupts will cause your PRI errors.
You can't fix broken graphics cards, nor fix the drivers.

> Another idea is that right now the system is only a single proc, but it is dual
> proc capable.  Would this somehow help if we added the second proc?  My
> thinking is it won't because it's a kernel module we are dealing with, and
> because of that I can't control the affinity of the driver (I was thinking at
> one point put X11 on 1 proc and Asterisk on the second, but it's not Asterisk
> that has the problem I don't think.)

I doubt it will make a difference. As I mention above, it ends up being
a kernel level problem and you probably can not get the drivers to
operate on different CPUs. It would just be better to spend the money on
a decent workstation and teach yourself how to manage the asterisk
machine from the separate workstation. 

> My final idea is that currently the system has an onboard 8mb ati graphic card
> that leaves almost all actual graphics processing to the CPU, could adding a
> better graphics card possibly help X use less cpu and not get in the way so
> much?

Not really, graphics sucks time away from a CPU no matter.

> Anyway, I know this isn't a supported setup, so if thats your answer don't
> bother replying, I'm know this will be a kludge/hack to get working (if I can
> get it working at all).  I'm just trying to do something that will be
> convienient for me and my users, there are other systems running on the server
> that I don't want to manage through the CLI, and the users don't know how to
> manage through the CLI, and there is no web management for them.

Ahh, so you are breaking many cardinal rules here. You have too much
stuff on your asterisk machine. You are begging for troubles. Move those
services off of the asterisk machine.

> Does anyone have success running X on an asterisk box that terminates a PRI?
> If so what hardware (video card, cpu, ram, mobo, etc)?
> 
> Thanks as I know this setup isn't supported, and I'm probably asking alot, don't
> think I'm just relying on the list for bizarre things, I've been trying various
> ways of doing this for the last 3 weeks, I can successfully run a vnc server on
> the box (without X running) and everything works, so for whatever reason it is
> getting a lower priority or something.  I really need to run GDM though as
> managing VNC passwords/usernames/desktop settings is quite cumbersome and if we
> can just get GDM running, we can use our ldap authentication server for logins
> to this box (which is what we were doing previously when we didn't have a PRI
> terminated on this box).

X and graphics drivers are big hogs on memory and CPU. VNC moves the
graphical portion over to the client machine. If you need GDM, why not
get X servers for your other machines and let GDM broadcast. This should
mean your X server run from whatever other machine should be able to be
configured to use the GDM and login. While the app will run on the
asterisk machine and be bad, the real CPU drain will not happen on the
asterisk machine. 

Just to recap, I am not flaming you no matter how it seemed in the
message.
-- 
Steven Critchfield <critch at basesys.com>