On Fri, 2 Jun 2000 Patrik Schindler wrote:
> At 19:30 Uhr +0000 01.06.2000, Konstantin Reznitsky wrote:
>
> >I have a strong feeling that your primary interfaces on the routing linux
> >boxes a set to the outside networks (NOT on the link in between)
>
> I know the term "primary interface" only with MarsNWE. Could you explain this
>further?
The first interface you define becomes the primary one. Afpd by default will bind to
the first zone on this interface unless
forced othewise in the startup of the afpd itself (something like
server_name@zone_name).
>
> > > asun recomended to upgrade to the latest by then kernel, but that did not help.
>
> He wrote that to me, too.
But with me it was RH-6.1 and I think kernel 2.2.12 :) awhile ago...
>
> > > I also had to force the afpds to the zones I wanted (not default). I'm still not
>sure
> > how all this will work in a chain with more then two routing linux boxes (luckely
> > I never needed this).
>
> Badly enough, it seems that multicasting (used for broadcasting nbp information)
>won't
really work (reliable) further away than one hop.
>
> I'll have a look at atalkd but I don't have much hope to get this fixed.
If you really can deal with the code of atalkd (I'm not good enough, sorry) I can give
you couple more tips:
1. Could get it working not only when both primary interfaces were set for the link,
but also with only one being on the link and
another one on the external network (this might be the way to get it working in the
longer chain);
2. I want to confirm what you said about routing, rtmp and zip are actually working
the problem seems to be in the propagation
of the nbp information;
3. My experiment was not really clean - one of the boxes had different types (3Com PCI
and ISA NE2000) of the network
adapters and that seemed to affect the situation as well ( in the mater of swapping
the network interfaces it was giving
different results).
4. The final layout of the network was a star consisting of 12 dual-homed routing
servers. Primary interfaces were set inside,
afpds bound to the zones on the outside, so they were showing up in their local zones
in the chooser. Worked good for about
6 months, then was replaced with appletalk routing gigabit switches (just for the
bandwidth reasones), servers are still there,
but as just servers.
So I never had chance to play with a chain of routers, but in the nearest future I
might have to, network is still growing
(that's why I'm interested in the solution).
Sorry, almost forgot! When the linux routing structure was still in place it also
included the Caman's Gator boxes with
ethernet connection to the local networks and their own localtalk segments. The
network was completely transparent from
any location. Gator box is not a linux router, but there we had a system with max 4(!)
hops.
I hope this will give you something...
> :wq! PoC
>
>
I'm also still interesed in the information on the appletalk routing. I asked this
question before and got contrary answers.
It has to do with redundant routing and equal cost multipath, is it doable with
appletalk? Any Ideas?
Konstantin
