Subject: Re: TCP/SACK support.
To: Jonathan Stone <>
From: Kentaro A. Kurahone <>
List: tech-kern
Date: 01/06/2005 22:33:12
On Thu, Jan 06, 2005 at 11:31:27AM -0800, Jonathan Stone wrote:
> After considerable thought, I'd strogly prefer to see NetBSD
> adopt/import one of the extant versions of of SACK, than to attempt to
> roll (and debug) our own in a tight timeframe.  I see's considerable
> benefit in sharing a common SACK implemetation, and a nontrivial
> downside in adding further divergence (I avoid the phrase "gratuitious
> divergence") to NetBSD's TCP stack.
> I have already spent some time back-porting portions of the FreeBSD-5
> TCP SACK code. There's also the DragonFly BSD SACK code. (and of
> course the Mathis &c PSU code).

Hmmm, ok.  I've only written the receiver path at (the easier side), from
what I understand writing the sender path is considerably more work, especially
in a way that is fast on links with large congestion windows[0].

I was going to look at how FreeBSD/PSC did the sender side and port/adapt as
neccecary.  No need to gratuitously reinvent the wheel afterall.

Kentaro A. Kurahone
SIGUSR1 Research and Development

[0]: I see references to Tom Kelly (Scalable TCP)'s work regarding a SACK fast
path that supposedly addresses a lot of these issues.