[aklug] Re: ACS <-> ACS VoIP Problems making me crazy crazy.

From: Shane Spencer <shane@bogomip.com>
Date: Fri Jul 10 2009 - 19:56:46 AKDT

OK. Here's what I've got so far.

1.) I removed all QoS from both sides and removed all routes except for those between sites, turned off IP forwarding on all boxes. Asterisk is being run on the gateway devices at Fairbanks and Wasilla, Anchorage has an Asterisk server with an external IP routed to over gigabit.

2.) Tested between Wasilla and Anchorage using iperf. no UDP packets had bad checksums. rate limited the UDP test to half of the available bandwidth.

3.) Tested between Wasilla and Anchorage using IAX2/Speex and an echo test. ~30% of the IAX/Codec packets had a bad checksum in the UDP header - call quality was good.

4.) Tested between Wasilla and Anchorage using IAX2/Speex over OpenVPN UDP tunnel mode. 0 of the UDP packets over the tunnel had a bad UDP checksum. ~30% of the OpenVPN UDP packets had a bad checksum in the UDP header - call quality was good.

I'm trying to duplicate the problem I have been having for months at the moment, any minute now.

*UPDATE* While trying to diagnose the issue, I turned off the RX/TX checksumming and TSO on my e1000 nics on all servers related to the calling. No UDP checksum issues. Sigh.. I'm starting to think that's just plain not important anymore :)

- Shane

On Fri, Jul 10, 2009 at 6:12 PM, Shane R. Spencer<shane@bogomip.com> wrote:
> Hey, Royce -
>
> Royce Williams wrote:
>> Hey, Shane -
>>
>> Shane R. Spencer wrote, on 7/10/2009 2:33 PM:
>>> My last resort will be IAX2 + Speex.. but for now I ask the jury.. What t=
>>> he heck is going on with my ACS services.  I have a BXB in Fairbanks
>>> and Anchorage and ACS DSL in Wasilla.  Fairbanks and Wasilla have several=
>>>  VoIP phones each.  Usually only one g.729a call at a time.
>>> Generally only 20msec of audio is sent per packet per channel.  Even with=
>>>  no site transfering data or browsing the web, I get calls that seem to
>>> have packet out of order or queueud packets.  Wherein 20msec is sent per =
>>> packet.. there seems to be a short burst of packets instead of a fluid
>>> stream.  Audio wise it seems like the playback is constantly trying to ca=
>>> tch up with either buffered or lost packets.  I need to do lower level
>>> diagnosticts.
>>
>> Speaking as a geek, and not as an ACS droid (yet, anyway), a couple of
>> questions:
>>
>> - Sounds like it was working better before; when did it go south?
>
> Approximately 4 weeks ago it went further south than it was before.  It started with intra-ACS connections then we started noticing poor call
> quality with our SIP peers in the lower 48.  I am setting up some testing in the lower 48 so I can check out RTP checksums.
>
>> - Have you sniffed the traffic to look for actual retransmits or other
>> 'smoking gun' evidence of drops?
>
> No retransmits since it's all RTP, however I see a LOT of RTP checksum issues.  I'm tempted to turn off RTP checksumming in iax.conf and watch
> the errors fly with codec_speex.so.  I'm not at that point yet.. still trying to wrangle a goat up for the obligatory sacrifice.
>
> Shane
>
>

-- Attached file included as plaintext by Ecartis --
-- File: signature.asc
-- Desc: OpenPGP digital signature

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.9 (GNU/Linux)
Comment: Use GnuPG with Firefox : http://getfiregpg.org (Version: 0.7.6)

iEYEARECAAYFAkpYDYAACgkQXK/vGhypreKgzQCeOWUxJWT4fbzIFrT/VNbgPLHu
OHUAoIA377XSgGyFzvUyKxcyJ6DnaoQd
=lUv9
-----END PGP SIGNATURE-----

---------
To unsubscribe, send email to <aklug-request@aklug.org>
with 'unsubscribe' in the message body.
Received on Fri Jul 10 19:57:09 2009

This archive was generated by hypermail 2.1.8 : Fri Jul 10 2009 - 19:57:09 AKDT