What is the best architecture to bridge to XMPP?

Question

If I have a separate system with its own concept of users and presence, what is the most appropriate architecture for creating a bridge to an XMPP server network? As far as I can tell there are three primary ways:

1. Act as a server. This creates one touchpoint, but I fear it has implications for compatibility, and potentially creates complexity in my system for emulating a server.

2. Act as a clients. This seems to imply that I need one connection per user in my system, which just isn't going to scale well.

3. I've heard of an XMPP gateway protocol, but it's unclear if this is any better than the client solution. I also can't tell if this is standard or not.

Any suggestions or tradeoffs would be appreciated. For example, would any of these solutions require running code inside the target XMPP server (not likely something I can do).

Answer 1

The XMPP gateway protocol you've heard of is most likely to do with transports. A transport is a server that connects to both a XMPP server and a non-XMPP server. By running a transport, I can use my Jabber client to talk to someone using, say, MSN Messenger.

A transport typically connects once to the remote network for each JID that it sees as online. That is, it's your option 2 in reverse. This is because there is no special relationship between the transport and the non-XMPP network; the transport is simply acting as a bunch of regular clients. For this to work, XMPP clients must first register with the transport, giving login credentials for the remote network, and allowing the transport to view their presence.

The only reason this has a chance of scaling better is that there can be many transports for the same remote network. For example, my Jabber server could run a transport to MSN, another Jabber server could run another one, and so on, each one providing connections for a different subset of XMPP users. While this spreads out the load on the Jabber side, and load balancing on your system may spread out the load as well, it still requires many connections between the two systems.

In your case, because (I assume) the non-XMPP side of things is cooperating, putting a XMPP server interface on the non-XMPP server is likely your best bet. That server interface is best suited for managing the mapping between XMPP JIDs and how that JID will appear on its own network, rather than forcing XMPP users to register and so on.

In case you haven't seen these, you might find them useful:

• http://www.jabber.org/jabber-for-geeks/technology-overview
• http://www.xmpp.org/protocols/
• http://www.xmpp.org/extensions/

Hope that helps.

Answer 2

I too am working on a similar system.

I am going with the gateway/component route. I have looked at several options and settled with this one.

The gateway is basically a component with the specific purpose of bridging Jabber/XMPP with another network. You will have to build most of the things you take for granted when using XMPP as a client. Stuff like roster control.

There is very little help online on the actual design and building of a component. Like the above answer I found that the xmpp protocols/extensions to be of help. The main ones being:

• Basic Client 2008
• Basic Server 2008
• Intermediate Client 2008
• Intermediate Server 2008

Reading through these will show you what XEPs you will be expected to be able to handle. Ignore the stuff that will be handled by the server that your component will be attched to.

It's a shame that Djabberd has such poor documentation as their system of "everything is a module" gave the possibility of backend of the server could interface directly to the other network. I made no headway on this.

Answer 3

One other approach is to work with your XMPP server vendor. Most have internal APIs that make injecting presence possible from third party applications. For example, Jabber XCP provides an API for this that's really easy to use.

(Disclosure: I work for Jabber, Inc, the company behind Jabber XCP)

Answer 4

There are basically two types of server to server (s2s) connections. The first is either called a gateway or a transport, but they're the same thing. This is probably the kind you're looking for. I couldn't find specific documentation for the non-XMPP side, but how XMPP thinks about doing translations to legacy servers is at http://xmpp.org/extensions/xep-0100.html. The second kind really isn't explained in any additional XEPs -- it's regular XMPP s2s connections. Look for "Server-to-Server Communication" in RFC 3920 or RFC 3920bis for the latest draft update.

Since you have your own users and presence on your server, and it's not XMPP, the concepts aren't going to map completely to the XMPP model. This is where the work of the transport comes in. You have to do the translation from your model to the XMPP model. While this is some work, you do get to make all the decisions.

Which brings us right to one of the key design choices -- you need to really decide which things you are going to map to XMPP from your service and what you aren't. These feature and use case descriptions will drive the overall structure. For example, is this like a transport to talk to AOL or MSN chat services? Then you'll need a way to map their equivalent of rosters, presence, and keep session information along with logins and passwords from your local users to the remote server. This is because your transport will need to pretend to be those users and will need to login for them.

Or, maybe you're just an s2s bridge to someone else's XMPP based chess game, so you don't need a login on the remote server, and can just act similarly to an email server and pass the information back and forth. (With normal s2s connections the only session that would be stored would be SASL authentication used with the remote server, but at the user level s2s just maintains the connection, and not the login session.)

Other factors are scalability and modularity on your end. You nailed some of the scalability concerns. Take a look at putting in multiple transports to balance the load. For modularity, see where you want to make decisions about what to do with each packet or action. For example, how do you handle and keep track of subscription data? You can put it on your transport, but then that makes using multiple transports harder. Or if you make that decision closer to your core server you can have simpler transports and use some common code if you need to talk to services other than XMPP. The trade off is a more complex core server with more vulnerability potential.