[Asterisk-Users] Re:(2) Henning G. Schulzrinne quote on IAX2 from von magazine

[John Todd]

[moved from -dev list due to non-dev topic content]

At 12:44 PM +0800 on 8/13/05, Steve Underwood wrote:
>Mike Taht wrote:
>
>>but hey, maybe the folk on this list understand where he's coming 
>>from and can explain why sip is better....
>
>He is one of originators of RTP and the main guy behind SIP. Of 
>course he thinks they are wonderful. The reality is they were both 
>poorly thought out, and people have been shoring them up ever since.
>
>RTP used endless ports for no good reason. Nothing was symmetric. 
>Bandwidth was considered of no consequence. It had no security, 
>which has only recently been grafted on with sRTP. People have put 
>massive effort into trying to live with or fix these things ever 
>since.
>
>SIP started from the notion that call control is fundamentally 
>simple, and H.323 was overly complex. SIP has been getting more 
>complex ever since, and is now as complex as H.323. It had no 
>security, and used an unreliable medium (UDP) for communication that 
>needs to be reliable. Now networks are changing this, through a 
>massive overhaul. To his credit Henning does accept that UDP for SIP 
>was a dumb idea.
>
>He does say some good things, like:
>
>"I consider the term "soft switch" a marketing term that, like its 
>cousin, "session border controller," seems to have no crisp 
>definition that everyone can agree on. Even the Soft Switch 
>Consortium has changed its name. The sooner we can drop the term 
>soft switch from our discussions, the sooner people and their 
>PowerPoint presentations will actually have to explain what their 
>architecture is, rather than hiding behind an ill-defined label."
>
>Of Skype "Most importantly, Skype got the "out-of-the-box" 
>experience right-most of the time, it just works, without 
>complicated configuration, even with NATs."
>
>and "One of the nicer things about Skype is that they avoid being 
>trapped in replicating the PSTN user appearance. Others have tried 
>to make their software applications look like a cell phone or desk 
>phone, which most often simply causes the software to inherit all 
>the usability limitations that ISDN and other feature-rich phones 
>had. A user interface stressing the buddy list and IM functionality 
>seems a much better fit. "
>
>and "Also, they provided higher-quality audio codecs rather than 
>feeling constrained by the notion that this would be wasted since 
>the PSTN only supports narrowband audio. The technology to do this 
>is fortunately readily available, both commercially and as 
>open-source codecs like Speex, and SIP-based soft clients such as 
>the new Yahoo Messenger and EyeBeam."
>
>He doesn't seem to really understand the strengths and weaknesses of 
>IAX. IAX has drawbacks, but none of the problems he lists actually 
>exist.
>
>Regards,
>Steve

   I won't debate Henning's comments on IAX2 other than to say it 
appears his comments reflect a less-than-full understanding of the 
protocol.

   I typically agree with Henning on many points when I read his RFCs 
and comments, but I suspect that on the topic of IAX2 he is going to 
be defensive as it is a protocol that has a distinctly pragmatic 
approach versus SIP which tries (too hard, IMHO) to be abstracted to 
allow any possible use.  It is difficult to understand all of the 
possible variations of SIP's protocol flexibility, which makes 
programming somewhat difficult and makes testing regiments VERY 
difficult when attempting to build core systems which have to digest 
all possible flavors of SIP.  The very nature of this complexity is 
why there is the hated "Session Border Controller" market in the 
first place.

   IAX2 is more easily understood due to it's more limited scope and 
regimented parsing, but it was not developed by committee.  It has 
less flexibility, but... do I need anything more?  Not typically. 
Voice, video, IM, HTML... sounds like a pretty full suite.  The jury 
is still out on the benefits of IAX2 vs. SIP as a protocol which can 
be more widely used.  Most SIP people of course will dismiss IAX2 
completely, as it challenges the group-think.  There are very few 
IAX2 developers, which is another struggling point for implementation 
of the protocol into commercial hardware.  There is no BSD-licensed 
version of an IAX2 stack, to my knowledge, which again cripples IAX 
integration into products.  It may be that SIP simply wins due to 
marketing, with IAX2's benefits and drawbacks never really being 
examined by most developers of applications.

  (Note to GPL-minded folks proposing a protocol: never, ever make the 
reference code GPL or LGPL.  Make it "public domain" or BSD-licensed. 
GPL impedes commercial implementation for embedded code.  Note I said 
"impedes" not "prevents".  But what adverb would you rather see 
describe your protocol uptake? "impedes", "prevents", "confuses", or 
"accelerates"?  You can make your application whatever license you 
want, but the _protocol_ stack implementation should be completely 
without restriction of uses.)

  I will strongly second Henning's comments on the fact that 
"out-of-the-box" is _everything_ to consumer uptake, and that it is 
extremely short-sighted to have a VoIP system that has as it's 
best-case codecs something that is equivalent to TDM telephony.  See 
my comments towards this end in the recent "G722" discussion on the 
-dev list.  Asterisk has high-quality codecs available to it - we 
(the community) need to implement them.  Speex can do what we need, 
now.  It's just a big programming effort to get it into Asterisk.

JT