💾 Archived View for gmi.noulin.net › mobileNews › 312.gmi captured on 2023-06-14 at 18:16:01. Gemini links have been rewritten to link to archived content

-=-=-=-=-=-=-

Congestion and all that.

2007-10-03 12:31:57

(Score:5, Interesting)

by Animats (122034) on Tuesday October 02, @08:22PM (#20831499)

(http://www.animats.com)

Having been around at the beginning, I should comment on this.

There are some fundamental problems with the way the Internet works, but

hardware has saved us from having to solve them. The biggest problem is that we

still can't deal effectively with congestion in the middle of a pure datagram

network. We know what to do out near the edges (look up "fair queuing", which I

invented), but in the middle, where there are too many flows and too little

transit delay, that doesn't work.

The practical solution to the problem has been cheap long-haul bandwidth in the

backbone of the network, with routers to match. Early users of the modern

Internet may remember the days when MAE-EAST and MAE-WEST would choke on

traffic and the whole backbone would start losing half the packets. That was

solved by cheap fibre optic links. Today, we have a network where the "last

mile" usually saturates before the backbone does. This is what makes the whole

thing work. But we never did get a good technical solution to that problem. We

have some good hacks: the congestion window in TCP and "Random Early Drop",

which together sort of work. At least where most of the traffic is TCP. We

still don't have equally effective ways of throttling UDP traffic.

Roberts is a virtual circuit guy. He founded Telenet, which was a virtual

circuit system. (I was recruited by Telenet when they had 13 employees, but

turned them down.) Telenet was a flop commercially; it didn't scale up well.

Telcos love virtual circuits, because they create connections they can bill.

And they keep trying to get virtual circuits into the network. X.25, ISDN, ATM,

and PPPoE are virtual circuit systems, and they all came from telcos. Roberts

is still pushing variations on his virtual circuit scheme.

There are continuing attempts to get some kind of billable virtual circuit

thing into the network, and those attempts consistently come from telcos. There

was a scheme tried for using multiple PPPoE connections over ADSL links to

provide multiple classes of service, with the good ones being more expensive.

That didn't fly. The whole "net neutrality" thing is about this. What telcos

really want is to be able to charge based on the "value to the consumer". The

wireless phone people do this, and cash in big - SMS messages cost more to send

than photos. The wireline telcos see themselves being cut out of the revenue

stream as video moves to the Internet. They want to create a place where they

can step on the hose and cut off the flow unless you pay them extra.

I wrote the classic RFC on this [faqs.org] too many years ago. Read the section

"Game Theoretic Aspects of Network Congestion". It's still valid. But, as I

said above, we don't have to solve the theoretical problem as long as throwing

cheap backbone bandwidth at it works. Cheap backbone bandwidth will continue to

be available unless some monopoly situation develops that prevents backbone

bandwidth from being provided near cost.