Skip site navigation (1)Skip section navigation (2)
Date:      Wed, 11 Feb 2004 13:35:17 +0000 (GMT)
From:      Richard Wendland <richard@starburst.demon.co.uk>
To:        silby@silby.com (Mike Silbersack)
Cc:        freebsd-net@freebsd.org
Subject:   Re: dummynet = local taffic > 100ms - help!
Message-ID:  <200402111335.NAA09040@starburst.demon.co.uk>
In-Reply-To: <20040211025940.J1798@odysseus.silby.com> from "Mike Silbersack" at Feb 11, 2004 03:01:25 AM

next in thread | previous in thread | raw e-mail | index | archive | help
> It breaks TCP Timestamp generation slightly, but that's not likely to
> break much of anything in practice.

Well, with HZ=10000 RFC1323 TCP connections will stop after 59.7 hours,
with HZ=100000 after 6 hours.  For those with long running TCP connections
(eg remote backup) that could be a big deal.  See 4.2.3 of RFC1323.

It does seem quite a few people want HZ>1000 so I think the time has
come to isolate the TCP timestamp option clock from the HZ value to
avoid this problem.  For now they should set net.inet.tcp.rfc1323=0 to
avoid breaking RFC1323.

Note this doesn't affect routed packets, only TCP connections to/from
that host.

Tom Pavel sent some patches to this list on 14 Jan 2004 that he has been
using to overcome this HZ/RFC1323 problem.

	Richard
-- 
Richard Wendland				richard@wendland.org.uk



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200402111335.NAA09040>