[ih] Global congestion collapse

David L. Mills mills at udel.edu
Mon Oct 4 13:48:45 PDT 2004


Michael,

The only incident I can think of when the Internet "collapsed" was in 
1987, and that did not involve the entire Internet as such. At the time 
the NSF phase-I backbone had six LSI-11 fuzzballs as routers, each 
connected to a hardware interface that did retransmissions when 
necessary. The backbone was connected at several points to the ARPAnet, 
most of which at 56/64 kbps. The fuzzbals were located at the five NSF 
supercomputer centers on various college campuses.

The fuzzballs ant 56-kbps backbone were hammered on occasion with 
massive amounts of traffic, not only from the supercomputers themselves, 
but with ordinary intercampus traffic. On a couple of occasions 
substantially all the packet buffers in the fuzzball network filled up 
and hardware-driven packet retransmissions displaced all new traffic. 
Eventually, the elephants that created the traffic went away and the 
network started working again.

The fuzzballs lasted until 1988 when they were eaten by IBM/MCI and the 
Phase II backbone at substantially higher speeds.

There is a paper in 1988 SIGCOMM if you want more details.

Dave



Michael Welzl wrote:

>Dear all,
>
>Does anybody here have stories about the Internet's congestion
>collapse(s) during the 80's? Some details would be great!
>
>Interestingly, RFC 896 describes congestion collapse as a
>stable state that is reached via unnecessary retransmits,
>whereas the probably more common description refers to a
>state where total network throughput decreases in response
>to increased input traffic (which can be caused by different
>link capacities along end2end paths).
>
>So, I wonder, what was it like? What are your experiences?
>When did folks first notice it?
>
>Cheers,
>Michael
>  
>




More information about the Internet-history mailing list