来自：
http://www.gnugk.org/sip.html
Will SIP take over and will H.323 die out ?
引用:
Probably not. While SIP is very strong in the consumer market, H.323 has a
very strong position in the telco market. Both for good reasons.
Here is a quote from Simon Horn on the GnuGk mailinglist in February 2006:
[...] Why haven't the major telcos switched to SIP? 2 reasons
1. LRQ There is no equivalent in SIP. You cannot deploy a large SIP server
farm network anywhere as efficiently as huge H.323 GK hierarchy
2. ISDN (Traditional Phone network). ISDN is H.320 and is designed for and
is far more compatible with H.323
However I will point out that H.323 does have better NAT traversal
capabilities (actually it does) and GNUGK NAT traversal method is far
better than anything capable in SIP and even within the ITU standard
H.460.18/19. There is also the ability to just simply port forward the
required ports in the router, this is more problematic is SIP.. The
reason, basically the design of the protocol, In SIP the receiving and
sending ports for RTP must be the same which means that port translation in
symmetric NATs (more than 50% of the market) is a major problem so STUN
becomes basically useless, you have to revert to ICE which requires up to
10 extra messages to traverse the NAT which does take some time to
negotiate and for business needs it becomes unusable. This is why your ITSP
insists you buy an expensive box to connect your phone to 
. The GNUGK
method uses 0 extra messages and instant.. Though I will point out that
some SIP servers such as SER do support a method similar to GNUGK but not
as efficient
Another major problem is securing SIP. I have spent quite a bit of time
looking at porting our security ideas to SIP and have come to one
conclusion, it can't be done without upgrading an entire Network to support
it. The basic problem is.
1. Lack of a dedicated End-to-End Calling channel (H.225.0 Signalling)
which makes it Impossible for 2 parties in a call to authenticate each
other directly. You need to rely on the servers and things like SIP
Identity (which requires 6 extra messages and can take up to 2 sec just to
identify the callers). In H.323 you just add a cryptoToken in the H.225
Setup message and your done. No extra messages, almost no extra delays)
2. Lack of a equivalent H.225 Call Proceeding Message which means its
impossible to do the encryption setup before the remote party rings. Which
means that you have up to 3 sec of 'Dead Air" when you pick the phone up.
For business this is unacceptable. They also cannot agree on which SRTP
method to use MiKey or sdesc so a lot of these secure phones/networks can
not interperate. In H.323 you can setup an encrypted call in under 400 ms
and it will be 100% interoperable with almost every piece of legacy H.323
equipment. In SIP you have to upgrade everything. If you pick the wrong
method MiKey or sdesc you may have to do it twice.
My point is from a programmers/implementation point of view H.323 is still
the only complete solution. Yes SIP is very good in it's box but outside it
it's not as good and for large scale business needs its not quite up to it.
[...] SIP is 10 years old, 5 years ago they were saying H.323 is finished SIP
will rule, well its 5 years later and H.323 is still here and the people
who were saying that are moving onto Asterix. My bet is that in another 5
years H.323 will still be around and the Asterix people will be moving onto
something else. Open Source SKYPE clone perhaps
Some light reading on SIP
http://corporaterat.blogspot.com ... ective-painful.html
http://www.dailypayload.com/2006/0130.html
Simon
推荐看下如下两篇文章：
Developer Perspective: The painful evolution of SIP
Celebrating SIP's 10th Birthday
