[TankGirl]

Interesting discussion, Jack & Mike! Some comments...

Jack:
the whole p2p experience is operationally different from napsters' server based protocols and that includes the amount of access to shared files and users' connections to other users thru the network (chat & hotlisting) .

however, since there is no finite upper limit on a p2ps' user base and the client continues to be refined, as is now being seen with morpheus, the day will come and soon when you'll be able to get Anything You Want, including alices' resturant, with just a keyboard and a click, making napster a distant though fond memory. as it already is for newbies .

Your point about p2p experience is excellent. The real attraction of p2p is in the experience of it. Morpheus, WinMX, Gnutella... they all can provide the basic experience of finding some stuff you are looking for and getting it easily. Fair enough, that is motivating enough to make people to organize themselves into p2p communities around the various clients. But there is so much more to it...

How many times I have heard comments like that of chickie-monkie's: "But I must say, I sooooooooooo miss the days of being able to use....Napster!" Yes, many people miss those days and for a good reason. The social tools of Napster together with its singular namespace were enough to provide the social thrill of being a permanent member in a global, never-sleeping online community that Napster grew to be. The post-Napster newbies may not even miss this experience as they have not known it... but they will come to know it, in time. Kazaa/Morpheus started out as a totally 'stuff-oriented' software in the spirit of Gnutella but the popular demand made FastTrack add browsing to it. As soon as you have browsing, the need for permanent (and working) hotlists becomes obvious. And so on... step by step the new generation of p2p is getting more social but it still hasn't reached the social intelligence level of the good old Napster.

Mike:
Doubt it Jack, they'd better come up with a way to do broader searches with less traffic. If they don't you'll still be searching only the supernodes closest to your connection. Seems MC2 put it out that's why they limit the number of hits allowed and there is a self limiting search boundary.

True unlimited P2P will return hits hours after the original request and cause havoc with the amount of traffic/data across the net. Even if each computer only quieries 2 computers after 20 quiery levels you're at 2 to the 20th power. That's a hell of a lot of data not to mention the cost to someone to get a file. Or do you think those search engines really go out and quiery every website every time you search for something?

Jack:
Allow me to repost a page from mcs' board where jaan, a contract programmer for fasttrack, attempts an explanation. there are now about 4000 active supernodes at any one time:

Q. The supernodes don't cache search results at all? So is it cached at the client then?

A. yes, the client caches the results for 10 seconds or so, just to protect the supernode.

Q. You said something previously about using different branches of the network for repeated searches.

A. well, they are not completely different. for example, the supernode you are connected to is always searched (because that's cheap). so are its neighbours, if i recall correctly. it's the third step that (effectively) searches a bunch of random supernodes on the net. however, as the third step is really expensive, it is performed only if the first steps yield less than N results, where N is about 10. in other words, the different branches only kick in if you search for really obscure stuff.

Q. Is Morpheus simply remembering it once it found that this user had these files?

A. yes, it's remembering the user/IP. so if the user did not have a static IP or was behind firewall, the chances of finding him are not so good.

- js.

"bunch of random supernodes" being the key here.

i note that i was never able to find rare songs parked on my other computer in over a year of using napster without first using hotlist.

The supernode approach has clearly been a good way to enhance the scalability of serverless p2p architectures. WinMX's WPNP applies a similar approach and so does Gnutella with its new reflector nodes. The fact that Morpheus can connect 1.5 million people even loosely with 4000 supernodes is not a bad achievement at all. The next thing to do is to make that connectivity tighter so that working hotlists and full search visibility can be provided over the whole network. Then there will be again the technical ground for something similar to the 'Napster experience' and even better.

How to proceed with the architecture? How to make the community of supernodes to work more efficiently so that searches can really reach the whole available content? One interesting way to enhance the search functionality would be to allow the supernodes and their 'clientele' to gradually self-organize into more content-conscious node clusters. For example, if you were regularly looking for reggae tunes, you would love to connect to a supernode hosting mostly reggae lovers. That way there would be much more relevant material available directly from your local neighborhood. And for the inevitable remote searches the supernodes could use their genre-awareness to form more effective links between each other. The next best targets for a locally failed reggae search would be other reggae-oriented supernodes etc. The idea is to allow the low-level nodes to 'migrate' between supernodes so that those with shared musical interests would get gradually closer and closer to each other. When you would join the network as a newbie, your searches would initially be 'blind shots' into your local supernode(s) and from there on to the wider supernode community. But should you repeatedly find something to download from a topologically distant supernode, this supernode could agree with your 'own' supernode to start hosting you so that you would be closer to people and content that you are interested in.

- tg