[asterisk-dev] possible zaptel problem with SMP and RAID1

Matt Fredrickson creslin at digium.com
Mon Jul 9 18:24:08 CDT 2007


----- "Steve Totaro" <stotaro at totarotechnologies.com> wrote:
> >> we have in production several SMP xeon and one core duo 2 servers
> >> always with sata sw RAID1
> >>
> >> all xeon servers works well on digium 4xE1 with full loads
> >>
> >> core 2 duo with asus server board using digium PCI 1xE1 were
> useless
> >> because it constantly 
> >> generates taps to the sound. On another board it works well. (we
> have
> >> tryed moving digium 
> >> into another PCI slots, no IRQ sharing etc...) Sangoma pci express
> >> solves this issue with the 
> >> same configuration.
> >>
> >> I dont believe that RAID1 is causing this. If not sure, try to not
> >> write anything to the disk. 

It's likely that the PCI interface on the new TE120P, as well as the TE4XX and TE2XX line of cards do not share that problem.  One big reason we redid the single span card was because its PCI interface behaves better in a wider range of environments.

> > In my case we don't use software RAID _EVER_. I despise software
> RAID.
> > Also in my case, not writing to disk is not an option. Our business
> is
> > about the recording of calls. The recording must go somewhere. Our
> > choice was to upgrade a few CPU intensive apps to distribute the
> work to
> > somewhere else. Also to implement a policy about interactive use of
> > machines doing production work.
> >
> > But you even point out that some boards are just not able to handle
> the
> > realtime workload. The time spent trying to figure out what it is
> in
> > such a complex system that make those boards less than ideal isn't
> worth
> > the amount of money to just buy a different board and try again.   
> >
> >   
> 
> Like a Sangoma board?

A Sangoma board is not a magic bullet that will fix buggy hardware drivers that it must coexist with on the system.  If there is indeed another driver on the system that is not peacefully coexisting with the zaptel drivers, then it is the source of the problems.

---
Matthew Fredrickson
Software/Firmware Engineer
Digium, Inc.




More information about the asterisk-dev mailing list