Comment by jallmann
11 years ago
> Has anyone got experience with both XMPP and SIP/H.323?
I've worked with all three. The H.323 vs SIP battle was fought a long time ago... for better or worse, SIP won. Any H.323 systems still in the wild can probably be considered "legacy."
That being said, while H.323 has more "batteries included," its call flow is accordingly more complex. SIP is not too bad, there is some subtlety to it, but the main issue is that you have to piece together dozens of RFCs for an interoperable system, not to speak of implementation-specific quirks.
XMPP is the more appropriate protocol in this context -- IM and group chat -- because it has had presence and ad-hoc messaging baked in from the start, including reasonably supported MUC. Ultimately, XMPP's issues are similar to SIP's -- the patchwork of XEPs is too unwieldy to work with. For an example, compare the multimedia bits (eg, Jingle) and the corresponding SIP session setup flow. The stigma of being a XML-based protocol designed in the 90's doesn't help, either.
The biggest reason SIP is sticking around, however, is that it has major buy-in from the telcos with IMS/4G. However, SIP is also a more appropriate protocol for traditional telephony. In that respect, comparing SIP and XMPP is a bit like comparing apples to oranges. With the appropriate extensions, one could behave like the other, but it wouldn't be their strength.
What are some of the reasons it's hard to design a good messaging protocol?
If you know your requirements up front, then it's not hard. In fact, it's a good exercise to do so -- probably one reason why messaging systems are a dime a dozen nowadays. What's hard is making a messaging system that is everything for everybody. If SIP or XMPP has taught us anything, it's that "extensibility" is a killer. Requirements evolve, and often it is perceived as easier or "cleaner" to start from scratch, rather than work around the idiosyncrasies of existing systems.
On the other hand, if you are trying to design a protocol to solve a domain specific problem (eg, in cryptography, multimedia, distributed systems, etc) then it becomes more of a research endeavor with all the associated pitfalls. In that case, there will always be the "shoulders of giants" to build off of. Unfortunately though, for every pioneering TextSecure, there are a dozen CryptoCats repeating the same mistakes of years past...
Sigh, sorry to hear that. I was hoping h.323 might work well for a personal call system with the option of ditching regular cell service. And just try and brute force most communication through h.323 :-/
It seems obvious multimedia calls is part of our future, and I refuse to let that be dictated by corporate lock-in -- be that Skype/Lync, Hangouts or whatever Facebook will launch when they enable videochat. And any videochat/conference needs text messaging/conversations too, so it seems like it would be nice to run everything through one server/service.
Which I had hoped made sense would be a h.323 server (along with open source clients+maybe a websocket/web-client bridge).
Guess it'll have to be SIP for phone (so I can ditch my regular cell service), maybe SIP for videochat (but I'm sceptical about the videoconferencing bit)... and XMPP for messaging.
As neither Facebook, Google nor Microsoft support either XMPP in general (making it feasible to set up a bridge server) anymore, nor federation (which would actually be useful), I suppose I'll just have to give up on the idea that there'll appear a chat network where I'll be able to reach most of my contacts with a Free/Open client that isn't forced to play catchup to corporate whims.
For a while, Facebook via XMPP worked (but not group chat), now it's back to sms. Which is annoying because sms is gated by the telcos, although it should be easy enough to make a xmpp<>sms gateway (eg: [s]) with a spare android phone. For personal use that wouldn't have to incur any sms fees (but would mean keeping a non-free cellular plan).
For others frustrated by the same, there's at least a "traditional" project to reverse-engineer working with FBs current chat: https://github.com/jgeboski/bitlbee-facebook
[s] http://projectmaxs.org/homepage/
[ed: Forgot to commend duckduckgo and fastmail on hosting proper XMPP with federation and registration (at least ddg[d]) enabled. That's one reason why XMPP is nice -- you can actually point non-technical users at a suitable client, and add a couple of screenshots how how to set it up with ddg -- and they can then federate/talk away (that is assuming you don't/can't/won't support registration on your own server -- in my case the main reason not to recommend people wanting to talk to me to register on my server, is that I wouldn't know what kind of "SLA" my own server would have -- so if they used XMPP for other things than talking to me, having an account with someone a little more dedicated to keeping a server open to the general public might be better. But the main point is that that just works with XMPP today. And it should've just worked with both Facebook chat and Google talk/hangouts/newnamehere. And it probably also should've worked with Microsoft accounts/skype/hotmail.
[d] https://duck.co/blog/post/2/using-pidgin-with-xmpp-jabber
]
> I was hoping h.323 might work well for a personal call system with the option of ditching regular cell service. And just try and brute force most communication through h.323
I guess you could try to run H.323 through a $protocol bridge, but I don't know why you would want to. Retrofitting modern features onto H.323 does not seem like fun. Because if SIP suffers from being too general, H.323 suffers from being way too specific in its architecture and technology requirements -- read through some of the associated specs, eg H.245, and it's not hard to see why SIP won. For the longest time, there was not even the notion of URI addressing...
> SIP for phone ... maybe SIP for videochat (but I'm sceptical
In SIP (and most well-designed protocols), there is no fundamental difference between an audio and a video call. It's just capability negotiation and RTP. Actual interoperability is the problem... and that is one of the reasons vendors like to create closed networks, so they can guarantee the experience end-to-end.
Regarding gating and lock-in: Part of it may be a desire for a proprietary walled garden [1], but a big reason is logistical as well -- one of the reasons Google disabled XMPP federation was because of abuse. This is the same reason you can't dial directly into the PSTN without going through a trusted trunk, and why unlicensed use of cellular frequencies (or any RF, really) is illegal.
[1] Actually, the reason the PSTN is federated is probably more an accident of history than anything else. If not for the AT&T monopoly (and subsequent breakup), people would have been complaining about siloed telephone services 50 years ago! See Tim Wu's The Master Switch for a nice treatment of this.
Right. Interesting that XMPP in many ways reimplemented some of the problems of both IRC and USENET wrt. federation.
> Retrofitting modern features onto H.323 does not seem like fun. Because if SIP suffers from being too general, H.323 suffers from being way too specific in its architecture and technology requirements -- read through some of the associated specs, eg H.245, and it's not hard to see why SIP won. For the longest time, there was not even the notion of URI addressing...
Sorry to hear that. My general thought was that for a personal system, it would be way lower barrier to use existing h.323 server/clients, than to make a whole new protocol -- and that's probably true.
But if it has been effectively abandoned for a while, SIP might be better -- even if what one might effectively end up with is "proprietary" SIP, as one picks a subset that works for a particular use-case.
Personally I have to complementary goals: The first, and most important, is to build something that allows group chat with archiving, hopefully integrated with optional audio/video and some kind of media sharing (be that wiki, or something more traditional like straight up sharing of files/documents) [But it it should be obvious that by uploading to a wiki with shared authentication/authorization, one would only need to share an url over chat].
The second goal would be to enable federation, at least among those with the know-how/interest in running their own nodes.
Oh, and goal zero would be to retain control, so self-host (although option to get it "on tap" as a service would be nice), and Free/Open software/protocols.
In the end, perhaps XMPP along with just video/audio via WebRTC turns out to be the least worst option. I'm a little worried that the "web-centric" solutions like Mozilla Hello/together.js etc is difficult to pair up with command line clients, bots and/or make accessible for those that need it (eg: braille terminals). I guess audio might be preferred to braille in the general case, even for asynchronous messages, but text is very flexible (eg: text-to-speech system exists).