The MPLS WG Archive

[Fred Baker <fred@cisco.com>]

At 11:36 AM 10/16/00 -0400, Curtis Villamizar wrote:
>ECN is not like FECN and BECN.  ECN is used for end-to-end congestion
>avoidance (TCP).  The ECN ECT (ECN capable transport) bit is used to
>indicate that the end systems are ECN capable and using ECN.  The CE
>(congestion experienced) bit is set by routers in place of a drop to
>indicate that a drop would have occurred.  TCP is expected to react as
>if a drop has ocurred, but the retransmission is avoided.

Actually, I would argue that it is a lot like FECN. The model for FECN 
(proposed by Goldstein as I recall) was that it would be received by a 
Frame Relay DTE, and be copied into an option in CLNS. CLNS would then 
deliver it to the receiving TP-4, which would implement the DEC Bit 
algorithm to attempt to adapt transmission rate to the knee (as opposed to 
the cliff) of the power curve.

In the ECN case, we have a network layer protocol rather than a link layer 
protocol, but again the idea is that congestion would be sensed en route 
and reported by the network layer (IP) to the receiving transport (TCP), 
which would take very similar actions to accomplish the same goal.