Re: [tcpm] Tuning TCP parameters for the 21st century

[[e-mail]]

>>> The SYN/SYN-ACK retransmission rate we measured turned out to be >> 1%
>>> in many cases (a bit surprising to us) hence the benefit.
>
> That points to some other problem going on. As a packet loss rate,
> that's quite high and causes problems elsewhere in TCP. It would be

It doesn't have to be a packet loss rate, maybe it was a
per-connection measurement. We also found this to occur
for > 0.5% of connections in our own measurement study:

Dragana Damjanovic, Philipp Gschwandtner, Michael Welzl: "Why is this Web
Page coming up so slow? Investigating the Loss of SYN Packets", IFIP/TC6
NETWORKING 2009, 11-15 May 2009, Aachen, Germany.
http://www.welzl.at/research/publications/networking2009.pdf


> useful to understand where the drops are and why. Retransmitting SYNs to
> a busy endpoint whose queue is overflowing doesn't help things, e.g.

I understand that now, too (thanks to Lachlan who
explained it to me), and hence apologize for having
been ignorant to this rather obvious issue in the
above paper.

- but what about retransmitting SYN/ACKs faster?
Since the other side already sent a SYN, it probably
isn't overloaded.

Cheers,
Michael