On: Why there is no Facebook killer: the death of the ...

[omginternets]

Not to mention the average developer. I just spent the past few days wrangling NAT-punching issues for a small p2p project of mine. To say "the first rule of p2p is don't do p2p" is an understatement.

The (very unfortunate) truth is that p2p is just more difficult than client/server.

[hvidgaard → omginternets]

That problem is partially solved by using a supernode overlay structure. It isn't particular elegant, and it relies on certain nodes being willing to work more for the network - but it can enable nodes behind a NAT to parcitipate relatively easily.

[omginternets → hvidgaard]

Sure, sure (and the tip is much appreciated), but this just goes to show that P2P is more difficult than client/server.

The point is that one has to have a very good reason to use a P2P architecture, as it tends to create problems.

[nxzero → omginternets]

Maybe it's me, but I'd assume the many-to-many is more complex that one-to-many, especially when laid on top of systems designed for one-to-many.

[omginternets → nxzero]

Good point. I was thinking of issues like peer discovery and NAT punching, but it's true that traffic topology changes as well.