[Asterisk-Users] wcfxs causing constant CPU spikes
Michael Welter
mike at introspect.com
Wed Dec 15 18:13:20 MST 2004
Jim Van Meggelen wrote:
>
> There's more to this than that, but my advice would be to not even
> bother until you've built a fresh kernel over the RedHat one. (check out
> the list archives -- you're not the only one with this problem).
>
> I'm right in the middle of something that may have a huge impact on all
> the IRQ latency issues everyone's been having. Let me know if you're
> interested in using your system as a guinea pig.
>
I'm running vanilla linux-2.6.9 over Fedora 2/3. I've replaced a
GigaByte MB/VIA chipset with an ASUS MB/nVIDIA and still get the spike.
I've purchased additional TDM22B boards. In fact, I've replaced every
bit of hardware several times over--from mother boards to Digium cards
to fax machines. And I've tried both i386 and x86_64 architectures.
I get the spikes with both the TDM22B and T100P cards. The spikes begin
when the Digium modules are loaded, and the CPU utilization goes to
solid 100% idle when the Digium modules are unloaded.
I use 'vmstat 1' to monitor CPU utilization. The system (not userspace)
utilization goes to 40% every three seconds. This spike is usually
within a one second interval but is sometimes spread across two
intervals (say 10% system utilization in the first interval, 30% in the
next interval). During the one second vmstat interval, the spikes occur
without a corresponding increase in the number of interrupts, and this
tells me an interrupt is being held for a very long time--200-400ms.
The end result is that the spandsp transmission is corrupted with
garbage printed at the fax machine.
I'm really discouraged at Digium's disinterest in this problem. I
understand they have limited resources with lots to do, but it only
takes a minute to reply to my email to say either they are aware or not
aware of the problem.
Thanks Jim, I'd be interested in assisting.
Mike
--
Michael Welter
Introspect Telephony Corp.
Denver, Colorado US
+1.303.674.2575
mike at introspect.com
www.introspect.com
More information about the asterisk-users
mailing list