[Iccrg] "Classic congestion collapse": be on the lookout for....

[Jim Gettys]

I have one credible report of a significant private network collapsing
with symptoms of "classic" 1980's flavour congestion collapse, along
with significant difficulty getting the network "up" again after collapse.

Please let me know if you know of other examples you can share.

Two trends are worrying:
    1) the retirement of Windows XP is making it much easier to saturate
networks in the first place.
    2) bufferbloat may also be playing a role, as it can destroy TCP
congestion avoidance. But we don't know, and can't get the data to know
right now either.  Buffers can hide in places we don't have AQM to
control the problem. And this problem has been growing...

Unfortunately, more information is probably unavailable about the
events, causes, and cures for the time being.  It's really hard to be
"first" with such a report publicly; much easier if you know you have
company.

For extra bonus points (not related to the above report), look for time
based congestion problems; we could end up hurting if there is self
synchronising behaviour going on.

I don't think I need to tell this audience about the RED manifesto, but
you might need to remind your friends.

But I'm worried, given the report,
                Jim Gettys
                Bell Labs