Posted by turbogeek
on March 2, 2004 at 2:12 PM PST
The P2P thought police are roaming the halls. Are they misguided or have they been reading too many white papers about Gnutella?
Ive had the same conversation with four different people this month. The conversation started when I was showing how a well-known ID can directly connect two peers. All four made the same observation: Doesnt that go against the idea that P2P is about creating ad-hoc networks and the dynamic discovery services and resources?
They have been lead to believe that P2P applications should not start with any information and should only be allowed to discover things in the network. They fell in love with the idea of discovering treasures by looking in hidden places. Hard to tell if it is wunderlust or just an old idea that stuck in thier minds from old assumptions. The P2P literature has not helped much either because most discussed P2P in terms of searching. This is not surprising, considering that most of the current P2P is oriented toward file sharing where 'all' data is 'discovered'.
The second thing that many in P2P talk about is discovery of a service as opposed to a file. This is what I call a inverse software agent. Software agents usually get passed from computer to computer to do work. The inverse agent is one that you discover some place else and use on your PC to interact with other agents. An example might be a datin service that you discover and then use to interact with others that are single. The inverse agent makes sense, but there is very little support for it. Yes, it is a good idea, but the software and security support is not there. It is thousands of times more efficient to just write and distribute an application rather than a plugin to extend your network capabilities. An application that genericly finds and loads applications is overkill unles it is very generic - at which point it then loses its value. Catch-22.
Making all applications start with a search at a higher level of framwork is also not a JXTA strength. JXTA is multipurpose. It is not partcularly useful as a search tool(at least without a specific type of search). It is not even optimized for searching. Yes, there are search mechanisms, but they are very limited.
There is one area of search optimization that the JXTA team is working on very hard: Finding another computer listening to a pipe address. Simply, a computer that wants to accept information or create a two-way conversation. Thats what most network applications do, so thats a reasonable area to optimize. That also means that even if we discover the application somehow, we still have the system to discover other computers in a P2P network.
But why throw away the purest view of ad-hoc networking? Well, I am not throwing it away. Just putting it back where it should belong. Applications should be applications - not a service.
Lets talk about the real crux of the argument, the well-known ID and how it is not a violation of a dynamic P2P network.
A Priori ID / Well-Known-ID
I was once taught a Latin phrase that has stuck with me but I have found difficult to follow all the time: Eschew Obfuscation. Translated it is simply, avoid complicated language. The problem with this sage advise is that when a new technology comes around we get ne buzzwords too. In JXTA, a well-known or a priori ID is an ID that is already known or can be created from known information. The best example of this is an email address. You know your email address, so does your email provider, and so do your friends. Because an email address is usually simple, like turbogeek@ cluck.com, it is simple for any application to require it as a way to identify you. In effect, your email address becomes an ID, plus it can be used for email.
But lets get back to confusing buzzwords for a moment. An ID in JXTA is a URN or Universal Resource Name. URNs can have many formats, but they are just a unique identifier. The JXTA form of the ID is found everywhere in JXTA from groups to pipes and to identify versions and types of code or data. IDs are everywhere in JXTA
But where is the discovery or ad-hoc cpability of this P2P system? Mainly in managing the network to discover peers and route messages. The network is created ad-hoc except for a few seeded peers that are used to discover the rest of the network. Beyond that, there is no requirement or preference that peers discover information. Thus for well-know identifiers, the network is crreated on the fly to connect peers via an identical ID.
Discovery in JXTA, as I have said, is a little inefficient for data searches. There are many methods for creating efficient searches. I describe one earlier in a prior blog on a six degrees of separation search (a Kevin Bacon network search). The six-degree search used well-known IDs created from email addresses to create connections between PCs. Using the well-known ID increased the efficiency of the network.
This finnaly brings me back to the guys that think a known ID is 'not' a way to build an ad-hoc network. You should have seen that the ID is indeed a transiant thing - at least by clients when using an email address. The connection between peers is also ad-hoc and discovered via pipe routing whenever you are mathing peers with identical ID. In addition the ID can be formed to represent any subject, service, location, thing, or person. All you need is agreement between peers looking for the representation. Sounds like well-known ID are the poster boy for P2P.
I Dream of JINI
Perhaps it is the dream of JINI applications that lingers in the P2P purist's minds? If just walking into a room with a WiFi PDA connects you to a local printer is cool, why can't you apply the same thing to P2P? For example, if I log into JXTA, shouldn't it automatically guess I am not looking for a printer?
P2P is not as straight forward as JINI. Realizing that there is a printer is not the same as discovering a service that does printing and thus might be useful. To look at this another way, you could create aJXTA dating service. If we apply a JINI model, a PDA owned by a single person would aoutomatically load the dating service and start looking for a mate. Might be nice, but really a bit of overkill. On the other hand, if I run a dating application enabled with JXTA, I can use data about your likes and dislikes to create an ID that can match you to someone else like you.
The bottom line? Feel free to use JXTA IDs created from known data. It is not against the spirit of P2P. The well-known ID is also an eligent way to solve many P2P problems. No one will call the P2P thought police. If they do, send them to me and I'll give them your Get out of Jail Free card.