[guelo]

The problem with a big adaptive p2p network is that it is close to impossible to test under real production loads.

[jrockway]

My guess is that the average supernode is on something like 6M/1.5M DSL. This is easy to simulate.

Now, if most of your nodes are on machines directly connected to 10GbE, then that's a problem. But most Skype nodes aren't. (I do imagine there are a few connected to 100M+ connections. But only a few.)

[ohashi]

I imagine quite a lot in countries like Sweden. I had 100Mbit in my apartment for cheaper than my shitty comcast which barely breaks 5.

[pyre]

They could have setup their 'mega-supernodes' running the new client in a production environment, so that if it crashed horribly, it wasn't a client machine that failed (or their entire network).

Upgrading (relatively) untested software network-wide on what is essentially their critical infrastructure is bad news. If this didn't happen now, it would happen sometime. It was just a question of when.