Re: [Tsvwg] Re: HighSpeed TCP for Large Congestion Windows

[Clearly I can't keep up very well.  So, sorry for the delayed
 response.] 

> Do you think it could be problematic to recover from a cwnd of
> thousand of packets?  What happen in a scenario (unrealistic?)
> where 1 HighSpeed TCP if filling a pipe of 10Gb/s, 100ms RTT, and
> then a second HS-TCP starts? How many packets are going to be
> lost?  Even if only 1/10 of cwnd packets are going to be lost,
> about 800 packets will be dropped.  Will sack be able to handle
> that? Probably not... and how many RTT will be necessary to
> successfully terminate a Fast (?) Recovery?  

I'm interested in knowing why you think SACK-based recovery will not
work effectively in this case?  I think it should work just fine in
the case when 10% of the packets are lost (regardless of how many
losses that happens to be).

(FYI: My mental model of SACK is draft-allman-tcp-sack-12.txt)

> And what will happen if a RTO expires? How many duplicated
> segments will be carried by the network?  Do you believe that this
> can be an issue?

Hopefully with SACK we will not send needless retransmits into the
network.  Or, at least not that many.

On the other hand, if you end up using Reno we have some recent
measurements that suggest about 1 out of every 3 retransmits is
needless.  So, given Reno's grossness this *might* be an issue with
huge windows.  

allman

--
Mark Allman -- NASA GRC/BBN -- http://roland.grc.nasa.gov/~mallman/
_______________________________________________
tsvwg mailing list
tsvwg@ietf.org
https://www1.ietf.org/mailman/listinfo/tsvwg