I didn't totally understand it, but this article linked from Slashdot is interesting, about a general internet slowdown problem: bufferbloat.

http://slashdot.org/story/11/01/07/0...inking-the-Net

I'm reading it now, I guess the point he's trying to make is that locally one can saturate a network link (because thinking of the larger picture if makes little odds if 100000 devices are buffering and then sending at random intervals or all sending continuously at lower rate - you should end up with the same amount of usage. I think....

So the slowdown will appear towards the end of the connections where the available bandwidth is lower.

/me continues reading to check he got the right end of the stick

The short story is that in the old days switches and network infrastructure only had very limited amounts of buffer space, so TCP could pretty easily rate limit itself as it was designed to do.

The problem these days, and as an end user you have pretty much no control over this, is that the same infrastructure now has significantly larger amounts of buffer space, so traffic moves in pretty much the same way traffic does through any major city, ebbs and flows depending on how busy things are, but the old way of limiting the flow no longer work so well between end points, that being you and wherever the data is coming from. The end result is that Gettys' is suggesting that this type of latency could become bothersome for near real time activities the bigger these buffer spaces become.