[e2e] Mixed ECN and Non-ECN traffic flows.
avg at kotovnik.com
Fri Oct 11 15:44:01 PDT 2002
Saad - this is a "tragedy of commons" situtation; if a small fraction of
people ignores rules they benefit. If majority ignores rules, all lose.
What I am saying is that tweaked TCP stacks and "accelerators" ignoring
congestion control were tried and found no commercial success whatsoever -
and not because end-users were knowledgeable or civic minded. They failed
simply because not a lot of people valued the (real) performance gains
high enough to purchase those devices or software packages.
BTW, ignoring congestion control does not mean just disabling cwnd code
in TCP -- you want to have FEC features to compensate for lost packets
and/or some form of large-window selective ACK. Obviously, cumulative
ACKs w/o congestion control would perform very very poorly.
The "UDP" applications were all real-time, and so didn't have problems
Also, historically speaking, the first widely deployed bulk traffic
application w/o congestion control was CU-See-Mee; this lead a coalition
of ISPs to filter CU-See-Mee traffic, thus forcing developers to fix it.
Been there, seen that :) Since then, the threat of boycott by network
operators (and general awareness of the congestion control issues) is what
is keeping streaming app developers in line.
On Fri, 11 Oct 2002, Saad Biaz wrote:
> This is what many believed: "if I ignore congestion, I get better
> performance.... This is not true". When you ignore congestion control, you
> will hurt also yourself. Why do you think many applications initially
> based on UDP (without congestion control) finally added congestion
> control... Do the following experiment: deactivate congestion control from
> your TCP (comment out instructions that reduce cwnd ), and enjoy the
> "better" performance.
More information about the end2end-interest