Why did Gnutella not die as predicted?

I am currently doing research on p2p technologies, specifically in the area of efficiency. When looking at articles on the topic of Gnutella's scalability, I came across one at http://www.tch.org/gnutella.html that outlines why Gnutella shouldn't be able to scale.

As this was written some time ago, and Gnutella has yet to go down im flames, I am curious as to what happened. Were the author's calculations wrong in the first place? Has the number of users not reached the point that such high bandwidth is required? If neither of the above is the case and Gnutella is indeed fairly scalable, what changes have occured that allow for this scalability?

If any of you can help with this it would be greatly appreciated. Thank you.

[CycloCide]

<font face="Verdana, Arial" size="2">Originally posted by codemachine:
I am currently doing research on p2p technologies, specifically in the area of efficiency. When looking at articles on the topic of Gnutella's scalability, I came across one at http://www.tch.org/gnutella.html that outlines why Gnutella shouldn't be able to scale.

As this was written some time ago, and Gnutella has yet to go down im flames, I am curious as to what happened. Were the author's calculations wrong in the first place? Has the number of users not reached the point that such high bandwidth is required? If neither of the above is the case and Gnutella is indeed fairly scalable, what changes have occured that allow for this scalability?

If any of you can help with this it would be greatly appreciated. Thank you.

</font>

Greg Bilson from LimeWire posted a response to that article at http://www.gnutellanews.com/article.html?id=4206

Thanks.

I was thinking that the problems with his calculations had mostly to do with assuming an unnecessarily large user base and connectivity, which do not reflect the current network of 20,000-30,000 users.

Of course, if 1,000,000 users were to log onto Gnutella with relatively large connectivity, the bandwidth issue would become a problem. But if connectivity were kept to the recommended levels and some other enhancements to the clients were made, I believe that even 1,000,000 users could be supported (although it would be a heavily fragmented 1M users).

This may be a bit ambitious with the current network and would probably break slow connections, but the use of Reflectors and connection management in clients may allow the network to scale much more than people imagine, even without introducing the concept of groups. I will have to look into this further.

Thanks for the help so far.

Brian Gallaway