My new TCP "fast recovery" code

[Michael Welzl < >]

Dear all,

I’m sharing my new timer-based “fast recovery” code - this could surely be 
improved and “cleaned” further, but I hope it’s ok as a starting point to 
take over.
As usual, I just put it all on my internal dedicated Rome-visit webpage:
http://heim.ifi.uio.no/michawe/romavisit/

I haven’t yet tested this “in depth” - I need to stress-test it to see how it 
works… but from some quick checks with different BDPs, queue lengths, losses 
here and there, it seems to work *at least* as good as normal TCP with SACK, 
and actually sometimes better  (more robust, somehow - e.g. it seems better 
at handling the massive losses at the end of slow start).  And that’s without 
SACK!

I’ll make some slides that explain this code and the underlying rationale a 
little for our meeting on Tuesday. When I have them ready, I’ll upload them 
to this page too so you can read them earlier if you want.

Cheers,
Michael