[asterisk-biz] Experimental/new VoIP rate search engine.

Sun Jan 4 21:04:42 CST 2009

DISCLAIMER: Much of the following rant assumes ideal (typical)
internet conditions.

On 1/4/09, Trixter aka Bret McDanel <trixter at 0xdecafbad.com> wrote:
>
>
> which is why I suggested looking at the media gateways, which are the
>  rtp endpoint addresses.

  Sure, but how many providers are actually capable of doing this:

- Billing paranoia (I have to be in the media stream)
- Supporting non-Tier 1 compatible codecs (popular with the Asterisk
community) such as GSM, Speex, iLBC, etc.
- Supporting IAX.  No Tier 1 (more like the equipment they use) will
take IAX.  If your "provider" is using IAX, your media is most likely
ultimately being proxied to SIP/RTP somewhere.  Best of luck to you.
- Asterisk based platforms.  Asterisk doesn't (yet) properly support
direct RTP setup.  Re-INVITEs are not appropriate for carrier trunks.
If your provider is using Asterisk, it is probably in the media path
for at least a little while (assuming the Re-INVITE goes well).

  This is not an indictment of Asterisk, IAX, or any of those codecs.
It's just the reality in this environment.

> But the problem with that is that recqual no matter how good it is can
>  only check based on where its run, it cannot check for any other network
>  that its not on.  And it can only detect issues at the time the test was
>  run, issues come and go, look at broadvoice, they either work well or
>  not at all and it floats from which gateway the call is going through at
>  any given time.  Further the particular route that is used at that
>  particular time the call is made can influence results.  Due to ratio
>  concerns some providers will dump traffic onto a link they would not
>  normally to balance it and keep that trunks routes costs lower, so the
>  sum of calls for that billing cycle can influence call quality.

  Thank you, I appreciate that.

  Recqual was designed to test carriers.  The carrier is the variable,
and the underlying machine, network, etc are the controls.  If I'm
testing multiple carriers from my recqual machine (which I often do) I
can tell you, conclusively, which carriers provide for better actual
call quality.  That's not definitively saying that one is *better* for
everyone, but from that machine, one IS better than the other.  No
question.

  One could even argue that with sufficiently good connectivity from
that machine carriers could be directly compared to one another, with
an inference that one could possibly be better than all the others for
everyone (or at least most people).  If one carrier proxies their
media in Dallas and another in NYC, the actual difference in call
quality should not matter that much (short of some major, often
temporary internet hiccup), even if the recqual machine is in Tampa
(or wherever).

  I know, I know.  This makes no sense, and I don't know what I am
talking about.  The reality is once you get past the last couple of
hops on each side it really doesn't matter that much.  I have run
recqual tests against multiple geographically diverse SBCs and
multiple Tier-1 carriers.  If the carrier is crap, they are crap.
Geography doesn't matter all that much.  If they are good and my setup
is good, recqual will (and does) report fairly consistent results
irrespective of geography (assuming everything else is consistent).
Every Tier-1 backbone provider's public internet SLA should provide
sufficient quality for VoIP.  You just need to get there and your VoIP
reseller does too.  Oh yeah, that SLA has to mean something too.

  What will matter, and what recqual will detect, is if their
connectivity is oversubscribed, their components overworked, or their
configuration idiotic.  Especially when testing over time.  This would
(and does) provide for somewhat relevant comparisons of one carrier to
the next, regardless of media proxying, etc.  If they are cheap they
are cheap and this will show.

  Carriers (you mention Broadvoice) throwing calls to random IP
addresses...  Shame on them.  Every Tier-1 I have dealt with lately
will give you an SBC for signaling and media.  That's it.  Some don't
even allow for "public internet" connectivity or will only configure
trunk groups so small it's basically worthless and rely on MPLS or
other QoS enabled connectivity, in which case you are *definitely*
proxying media to your public internet customers.  It's up to the
reselling carrier to properly leverage this.  If they don't they are
doing their customers a disservice and this will show both in recqual
tests and complaints from end users.

  If you are "lucky" (I hate to use that term because those of us that
deal with these carriers should never be considered "lucky") you will
be given multiple, redundant, geographically diverse SBCs to originate
and terminate calls; complete with consistent, verifiable RTP endpoint
addresses.  Luckily that provider will also be very well peered (most
are) and you will be in a situation where you can provide for
consistently good call quality as long as you know what you are doing.

  If you look at recqual it was designed to watch for the actual RTP
endpoint address in the SDP.  This was not an accident.

<sales pitch>

  Star2Star, for example...  I don't have the exact numbers of the
last report but at least %90 of our customer's endpoints are within
50ms or less average RTT from the nearest Tier 1 media gateway/SBC
(you can probably guess who this is).  Most are far better than that.
We provision the endpoint, it handles media, and it's always on the
edge of the customer's network.  We can (and do) constantly monitor
this situation to adjust our provider agreements, outbound/inbound
routing decisions, and upstream IP trunk group configurations
accordingly.

</sales pitch>

>  As such using tools like that can be highly misleading as to quality,
>  why I just suggested looking at how well connected the network is where
>  the proxies and media gateways are located and suggested not doing a
>  quality ranking based on the service since its the sum of all the parts
>  at the time that the test was performed and unless you do this from many
>  origination points over an extended duration, and is kept current the
>  results can be misleading at best.

  Proxy the media once, as "carriers" often do (see reasons above) and
all of a sudden this doesn't matter.  They could have the best public
facing connectivity in the world, proxy their media and hand it off to
the next carrier over an oversubscribed Cogent link.  You just don't
know.  Actual, experienced audio quality is the *only* thing that
tells the full story.

  Not to mention, let's take an architecture like Level(3) Vector.
How many resellers properly leverage that with the proper components?
I might be pessimistic again, but I doubt very many do.

> that is a start but the tests have to be ongoing, it may work well
>  initially, then a surge of customers start using the provider, and their
>  quality degrades until they increase capacity.  Silent back end route
>  changes can have a similar effect, and you have to have different
>  destination numbers all over as they would also be influenced by the
>  routes that provider uses.  What reasonable provider does not LCR at the
>  very least?

  Agreed.  Those of us that have been around long enough (certainly
you and I, along with many of the others in this discussion) have seen
the ups and downs of various resellers.

  It is very, very difficult to LCR *properly* while still having an
ideal network architecture (direct RTP setup, for example) while
providing somewhat consistent call quality.  One provider uses Sonus,
another Lucent, another Cisco.  How do your customers all deal with
even the various intricacies in RTP implementations alone?  Each of
the vendors I just listed has a myriad of known RTP issues, whether it
be timestamps, sequence numbers, or RFC2833 issues.  I know you know
what I'm talking about here ;).  Last I checked the various major SBC
manufacturers were sitting pretty.  Business is booming.

>  I am just saying that quality measurements have to be done many times in
>  an ongoing effort per provider, with multiple origination and
>  destination points, and all that and even then it can be misleading for
>  some customers since they may not be as well connected as any of the
>  test boxes for a given provider but are better connected for others.

  In a perfect world, yes but I still maintain some *base* numbers can
be obtained fairly easily from one point of measurement.

  While this rant may read contradictory it's probably from lack of
clear definition:

- Carrier (Qwest? Broadvoice?)
- Tier-1 (IP? VoIP?)

  I (somewhat foolishly) interchanged these terms when describing
various entities.  For the sake of clarity and argument I could go
back and remedy this but I've already spent enough time on this at the
moment.

-- 
Kristian Kielhofner
http://blog.krisk.org
http://www.submityoursip.com
http://www.astlinux.org
http://www.star2star.com