[e2e] Can feedback be generated more fast in ECN?

[Eric A. Hall]

Hi Jon,

> oh, and 30% Loss means you either have under bufered routers, or
> severe overload where there's either lotsa unresponsive
>  streaming traffic , or you are in a severe underprovisioned case
> (yes, i know this happens in developing regions more (e.g. like
> england to us links in the recent past:-))

Not all congestion is poor planning or under-provisioning. Look at what
happened just a few days ago. Latency rates to sites all over the Internet
increased by several multiples within five minutes of CNN announcing the
Napster verdict, with everybody rushing to d/l stuff while the servers
were still up. I saw ping times in the 200ms range when I normally get
10-20ms from the same destinations.

Popular sites come and go (the dancing babies one week, dancing gophers
the next), and the traffic pattern shifts, overworking the designed-in
bottlenecks. Congestion pops up, settles down, moves somewhere else and
the cycle repeats.

Yes a lot of congestion is a result of provisioning problems. The exchange
point congestion patterns I am seeing is exactly that. Microsoft's two-day
outage from a couple of weeks ago was a direct result of man-made
congestion. While SQ and ECN wouldn't have helped much in the latter case,
SQ would work well in the former to keep the senders throttled until the
bottlenecks are widened.

As for the "developing nations" bit, I kind of assume that the Internet
*is* a developing nation. :p

-- 
Eric A. Hall                                        http://www.ehsco.com/
Internet Core Protocols          http://www.oreilly.com/catalog/coreprot/