Subject: Re: silo and packets: overflowing too-bigness
To: None <port-mac68k@NetBSD.ORG>
From: Bruce Anderson <brucea@wavefront.com>
List: port-mac68k
Date: 07/28/1998 17:27:00
On Tue, Jul 28, 1998 11:40 AM, Frederick Bruckman
<mailto:fredb@fb.sa.enteract.com> wrote:
> On 27 Jul 1998, Bruce Anderson wrote:
>
> > Then add the line:
> > sysctl -w net.inet.tcp.rfc1323=0
> > to your /etc/netstart.local
>
> What does that do?
---------------------------------------------------------------------
>From netbsd-help@netbsd.org
Forwarded message:
> From alex@vsys.com Wed Jul 8 19:04:41 1998
> Message-ID: <01BDAA60.F4AE7C20.alex@vsys.com>
> From: Alex Barclay <alex@vsys.com>
> To: "'NetBSD'" <netbsd@hgb-leipzig.de>
> Subject: RE: Slow connection between NetBSD and Sun Solaris
> Date: Wed, 8 Jul 1998 11:10:01 -0600
> Organization: Vsys
> X-Mailer: Microsoft Internet E-mail/MAPI - 8.0.0.4211
> MIME-Version: 1.0
> Content-Type: text/plain; charset="us-ascii"
> Content-Transfer-Encoding: 7bit
>
> On Wednesday, July 08, 1998 6:38 AM, NetBSD [SMTP:netbsd@hgb-leipzig.de]
wrote:
> > > Can you also try changing the sysctl variable:
> > >
> > > net.inet.tcp.mss_ifmtu
> > >
> > > to 1?
> > I tried, but nothing changed.
> >
> >
> > > I guess a tcpdump (or two from both ends) wouldn't hurt.
>
> I've had a quick look an it seems that there is some retransmission
occuring.
>
> There are 2 cases in which I have seen this... Both when there is a ppp
link
> involved..
In this case, there is a PPP link (2xNovell ISDN router) involved.
>
> (I guess that ISDN may not help here)
>
> OK here we go
>
> 1) The MSS negotiation gets screwed up. This is because the mtu of the
> interface
> is larger than the mtu of the ppp link. You'll notice the Don't
Fragment
> being set
> on the TCP stream. Se here are some suggestions if it is this
>
> On the netbsd side
>
> sysctl -w net.inet.ip.mtudisc = 0
>
> and
>
> sysctl -w net.inet.tcp.mssdflt = 128 # A bit small but should be way
less
> than the ISDN bit!
Nothing changed.
>
> 2) If Van-Jacobson header compression is in use then packets using the
rfc1323
> extensions to tcp can
> become corrupted. rfc1323 is designed to improve performance on high
> bandwidth x delay networks.
> So try to turn it off (I don't even know if Sun supports this?)
>
> On the netbsd side
>
> sysctl -w net.inet.tcp.rfc1323 = 0
>
Applying sysctl -w net.inet.tcp.rfc1323=0 solved the problem completely.
Thank to all answered me.
Best regards
Uwe