Peer-to-peer revolution? We'll see

In response to the report written by John Borland, "Super-powered peer-to-peer":

Yup. Revolutionary.

We tried a NeoNet link, and its downloads didn't work.

We tried an obscure search, and the only results that came back were from the Gnutella network and specifically from LimeWire clients.

State-of-the-art Gnutella technology (as embodied in LimeWire and partially in BearShare) has solved the searching problem in reasonably large networks (650K+ and growing). We have also solved a lot of the problems that they have not addressed, such as sources that don't work and improved download reliability. LimeWire has encouraged Morpheus all year to improve the out-of-date implementation of Gnutella that it has been using.

Distributed hash tables (DHTs) have a bad habit of holding stale data and blowing up when the network gets too large or the transients get too large. These are both issues that peer to peer is supposed to be able to handle. It is highly unlikely that a DHT will scale. The researchers from Stanford University who study P2P hate DHTs.

Petar Maymounkov, the inventor of Kademlia (a DHT that eDonkey/eMule try to use), now says DHTs are not only the wrong technology but also the wrong metaphor. He's now looking at routing based on trust--which is more where Gnutella is headed, in my eyes.

The intelligent routing that LimeWire invented minimizes the bandwidth used in searching while returning fresh and responsive results to the user. Our new firewall-to-firewall transfer capability, which we have in beta, will revolutionize the way peer-to-peer clients deal with the growing problem of firewalled hosts.

DHTs are the wrong way to solve a problem that doesn't exist. Meanwhile, solutions that do work are not being given the attention that they deserve. Anyway, it will be interesting to see how this plays out.

Greg Bildson
CTO, LimeWire
New York

 

More Letters | More Perspectives