On Sat, Apr 23, 2005 at 09:27:28AM -0700, Keith Fetterman wrote:
> Tom Trebisky wrote:
> >I thought I would followup post since I have been having a similar problem
> >on my home system, and MAYBE have a clue for a solution.
> >
> > ...
> >My home system has an intel etherpro 100 NIC, and connects through some
> >dogmeat ethernet switch to a Cisco 678 DSL router. These are the details,
> >though I don't think any are particularly relevant.
> >
> >Last night I added the following entry to my /etc/sysconfig/network
> >file:
> >
> >IPV6_NETWORKING=no
> > ...
> >This was right after experiencing a stall using scp to copy a big
> >file to my home system. After this, ... things seemed much better.
>
> I will check because you might have hit it. One difference between FC3,
> which is the 2.6 kernel, and my RHE3 systems, which is the RH modified
> 2.4 kernel, might be the IPv6 settings. In the RHE3 they may be
> different or turned off by default, where as they may be turned on by
> default in FC3. I will check and retest.
I am now somewhat less enthusiastic about this, although it -seemed- to
work. Apparently the variable to set is:
NETWORKING_IPV6=no (not IPV6_NETWORKING) -- I just reported verbatim what
some other poster advised, but a look at /etc/rc.d/init.d/network shows
that NETWORKING_IPV6 is the relevant variable, so it isn't clear how setting
IPV6_NETWORKING could do anything at all -- just a placebo ... sigh.
And reading the posts about slow network and IPv6 issues indicate this is
primarily a DNS thing, not something that would slow down or stall an
existing TCP connection (which is what I am seeing). see:
http://weblogs.mozillazine.org/darin/archives/005753.html
To really kill ipv6 it is recommended that the following be added to
/etc/modprobe.conf and then REBOOT (just restarting the network won't
yank the ipv6 modules out of the kernel).
alias net-pf-10 off
alias ipv6 off
Then to check if ipv6 is really gone, do ifconfig -a and verify
that no ipv6 address is assigned to your network interface.
This is as much as I have learned. There definitely is something amiss
as far as TCP connections stalling now and then, and it really is a
kick in the head when it happens in the middle of editing a long email
like this one (knock on wood). At first I suspected my ISP (now changed
for other reasons), then I suspected my wireless access point/switch
(now bypassed) -- but it seems to trace to fedora core 3, a fellow at
work reports that he is seeing similar incidents in the same timeframe
(i.e. as of the 2.6 kernel). It would be better if the connection would
timeout and fail rather than stall and deadlock.
Tom
--
Tom Trebisky
MMT Observatory
University of Arizona -- Tucson
tom@xxxxxxxx
