[MLB-WIRELESS] IP Allocations

[Dan Flett]

> Hmm yes at one stage a 172.16.x.x/32 was for backbone.  Of course this
> technically doesn't work as a /32 is useless and cannot route at all.
> (real
> brains trust came up with that one :-p)

Cool, then no-one should be using it. :)  If they are, I'd like to hear
from them!
 
> I'm a bit iffy on a generic config file.  For starters, a router-id is
not
> required to as OSPF will just choose an interface to use at the
router-id
> if
> none is specified.

Funny, just before I read your email I read a Quagga tutorial site
saying the exact same thing.  So one less thing people would need to
edit.
 
> They will need to change their own interface details, and their ospf
> network/neighbour details.

Basically what I've written so far defines all of the Melbourne Wireless
OSPF areas in the one file, and there are no interface definitions
anywhere.  I've done a day's worth of testing in RGSouthern with it (4
nodes) and it seems to work.  Quagga doesn't seem to mind if you don't
define your interfaces, it finds them itself. If you have an area or
areas defined it puts the appropriate interface in the appropriate area.
Quagga also doesn't seem to mind if you have an area defined that none
of your interfaces is in.  As far as I can tell the only major thing you
miss out on by not defining your interfaces is defining them as
broadcast, multicast or ptp.

> These lines are essentially all the configs consists of anyway so they
> pretty much end up creating a config file.  What we could create is a
> script
> where they put in their interface details and it spits out the basic
> config
> files.  In fact I think this may already exist somewhere online.

Quagga fills in the gaps in the config files when you run it - if you
run Quagga with an almost-bare config file and log into it's
command-interface VTY, you can view a version of it's running config -
what it would write to it's config file if you choose do so - your
interfaces are added automatically.

> Also, be cautious with areas.  As it turns out, people may need to
change
> the area they are actually in as the network grows so as not to break
area
> 0.  At this stage people should be using the area number for their
region
> until such time as we create an area 0 once a backbone appears.  At
this
> point the network will need to be juggled in these areas potentially
> splitting regions into subregions purely for routing protocol
purposes.

The idea with the file I'm making up is if people use it and use the IP
address range given to them by LocFinder, they will by default be put
into the correct area by Quagga.  My ospfd.conf doesn't define area 0 at
all, but it does define each Region Group's /20 supernet and associated
area number.

The beauty of the way the Melbourne Wireless network is structured and
it's use of OSPF is it's scalability.  It's a lot more scalable than the
mesh routing protocols out there such as AODV and OLSR.  But for us
OSPF's drawback is it is higher maintenance than the mesh protocols.
Our network runs entirely on volunteers of varying degrees of technical
skill.  Our network has no paid IT support staff, and many potential
node-owners are put off by the degree of technical knowledge required to
set up a full-blown, multi-interface, OSPF routing node.

As far as I know, there are no user-friendly front-ends for Quagga or
any other OSPF implementation anywhere.  I'm investigating ways to make
it easier for potential node-owners to participate in the network.  So
I'm looking to find out if it's possible to run OSPF with a generic
config file for the MW network.  I'm hoping we can configure run OSPF in
a stable and robust manner but not require the end user to get heavily
involved in it's setup.

Dan

To unsubscribe: send mail to majordomo at wireless.org.au
with "unsubscribe melbwireless" in the body of the message