Re: [tcpm] initial RTO (was Re: Tuning TCP parameters for the 21st century)

[Mark Allman <[e-mail]>]

> the rate distribution is far from even. 

Right.  That's clear from both your slides and a ton of previous
empirical observation.  I was merely using an average of sorts from your
slides as an illustration.

A couple more things:

  - Even if the loss rate is 10% that's .1*.1 or 1% instead of .02*.02.
    So, we're not helping a stunningly large number of connections.

  - Moreover, if we're talking about correlated loss rates across RTOs
    then I think you're arguing for aggressiveness in the face of
    persistent congestion to help a quite small fraction of
    connections.  That doesn't quite seem useful.

> As I mentioned in my previous email the try-N can diverge into
> different flavors.

It seems that we don't quite have the same notion of what we're talking
about.  Could you sketch something concrete in an email to the list?

Thanks!

allman