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

[Zhang Miao]

Hi, Neil Spring
    I can't understand how the conclusion "the ECN is more reliable" can be deduced
from the argument you provied below.
(1) If 8 packets in a window can trigger 4 ACKs with ECN, it is likely that 8 packet
   can trigger at least 4 SQs. Why did you use (0.05)^4 for ECN and use 5% for SQ?
   (Is there any limit for SQ? Did I make a mistake here?)
(2) The loss rate for ECN and SQ is different. The total path for ECN is longer than
   that for SQ. So if you assume a loss rate of 5% on the return path for SQ, you 
   must assure a higher rate for ECN.

  What I said above does not mean that I stick to SQ. From the discussion, I understand
more to the trade-off between ECN and SQ. In my understanding, 2 arguments are strong
enough for ECN:
(1) consideration for congestion control in multicast
    I don't have much experience on this topic.
    It is very reasonable for the receiver-based scheme.
    I would be grateful if anyone could provide a reference to me.
(2) the overhead of SQ processing at the router
    Though it is a big problem now, I wonder whether there is space for optimization.
    If SQ scheme only has this problem, it is worth to do more work in the router.
    The global performance of the network is more important, isn't it?
another argument is also very reasonable:
(3) whether the benefit is big enough if feedback is sent faster.
    It is preferred to determine whether the benefit is worth the price we have to pay
    before we do it. I'm not sure with this question.
    What I want to say here is, we should consider a link of multiple flows(perhaps
    thousands of) instead of one flow. A little modification to the algorithm may have
    a large result with the combined power of a lot of flows.

  Many people have discussed on the quality and congestion situation of the return path.