Subject: Re: Recent NetBSD network code improvements
To: Jason Thorpe <firstname.lastname@example.org>
From: Mark Allman <email@example.com>
Date: 05/01/1998 14:12:14
> [ Quick note - this message was originally CC'd to netbsd-announce, but
> the appropriate list for this sort of discussion is tech-net; I
> have changed the CC appropriately. --thorpej ]
Sorry... I actually didn't mean to CC anyone. ;)
> The default value is still 1 segment. To be quite honest, I was
> still somewhat unclear on what the `decision' was in LA... do we
> go with a 2 segment IW, with a 4 segment IW, or with Sally's ~4K
I think the appropriate thing to do is to use a 1 segment initial
window right now. In LA there were a small number (< 5, by my
count) of objectors to using the larger initial window. And, since
then there has been a little noise on the mailing list, but I don't
*think* that will hold things up. I expect that the I-D will be
forwarded to the IESG fairly soon. But, I am purposely a bit out of
the loop on that decision (since I am an author on the I-D am
setting aside my co-chair hat on this issue).
Assuming the I-D makes it to RFC status, I think your choices for IW
-the equation given in the draft
-something that is ALWAYS more conservative than the equation
in the draft (like IW=2, for instance)
The draft does not advocate just using 4 segment initial windows.
It advocates using 4 segment initial windows in *certain
circumstances* (i.e., when the MSS is small enough). (Same goes for
an initial window of 3 segments). So, I would cross off "4 segment
IW" from your list above.
While I don't think there is a large objection to 3 and 4 segment
initial windows, I think there is next to no objection to a 2
segment initial window. So, if you were to make that the default, I
doubt anyone would yell too loud. (And, as you noted a lot of
systems use 2 segment initial windows already).
I would personally be conservative and use the 1 segment initial
window for now. That being said, I think having the knob is
important, so that people can start playing with this on a wider
scale. We may end up seeing some unanticipated behavior.
[Sort of related question: Have you guys fixed the bug that caused
the 2 segment initial window?]