Re: tcpip stops working after some time
On Sat, 27 Dec 2008 16:16:00 +0200
Micha Feigin <firstname.lastname@example.org> wrote:
> On Wed, 24 Dec 2008 13:49:53 -0900
> Ken Irving <email@example.com> wrote:
> > On Wed, Dec 24, 2008 at 02:16:26PM -0800, firstname.lastname@example.org wrote:
> > > >From: email@example.com
> > > >>On Wed, Dec 24, 2008 at 10:25:07PM +0200, Micha Feigin wrote:
> > > >>> Ken Irving <firstname.lastname@example.org> wrote:
> > ...
> > > >>> >
> > > >>> > A too-big MTU setting can give odd results in some cases.
> > > >>> >
> > > >>>
> > > >>> but how would it change on a running system?
> > > >>
> > > >>It probably wouldn't, but maybe an upstream router could have. I'm
> > > >>just tossing out the idea in the off chance it might apply, given
> > > >>that some packets go quickly and others don't.
> > >
> It's wierd but it seems at the moment like the MTU has something to do with
> it, and on the local computer, nothing along the way, and it's not too high,
> it actually dropped too low for some strange reason.
> It's down at 576. I raised it to 1500 and the link
> is running tcp again. but really comes and goes and every time the link goes
> down the mtu drops again. Don't know if it's wicd or not yet.
It looks like dbus goes crazy at some point and starts setting the mtu to 576
once the connection is made. Killing just about anything possible and setting
the IP by hand makes most sites work, although not all of them for some reason.
fixing the mtu also helps some sites (less than the more drastic option) but
every time it goes down and up (wicd) it is reset.
What can make dbus mess up the network connection? (both ethernet and wireless
if it makes a difference).
> What can make the mtu drop on a running system?
> > > AFAIK the MTU is set by the initiator of the IP connection and any
> > > intermediate router can't change it.
> > MTU is a property of a link, not of packets; I was suggesting that
> > perhaps a change occurred in the settings of a router through which
> > packets are going. This is probably much ado about nothing, but it
> > is at least easy to test, e.g., by bringing up the interface with a
> > lower MTU, and seeing if the problem goes away. Years ago I had to
> > mess with this to accomodate a (possibly broken) DSL or ppp interface.
> > Some of the symptoms described in this thread sounded familiar, e.g.,
> > sometimes things work great, other times slow or not at all.
> > Ken