Re: [tcpm] TCP tuning

[Mark Allman <[e-mail]>]

[These are re-sends from long ago... they never made it to the list.
 But, that seems to have been worked out, now.  --allman]

> I think you need a negotiation for
> opt-in rather than opt-out.

If you want to negotiate an initial window then define a riff on
Quick-Start that doesn't require help from the routers.

But, that feels like too much here ...

  - If the receiver is the problem then even overwhelming it with a
    larger initial burst doesn't hurt the shared network much.

  - On the order of 10 packets it seems hard to believe that we're
    really going to overwhelm any even moderately recent end system.

  - If it isn't the end system we're worried about then the receiver is
    acting as a proxy for something in the middle.  And, so ...

      + How good will the receiver be able to achieve this?

      + If it is really a problem then you probably want something like
        Quick-Start to actually engage the path in the decision.

I helped write Quick-Start so don't get me wrong here.... perhaps there
is a need to get into all this mechanism if you want to send Really
Fast.  But, it seems that we should be able to have some initial value
that can be used without consent that will not cause issues high quite
high probability.  And, 'network inflation' would likely mean that
number should increase over time.

allman