[MLB-WIRELESS] Re: [RGSouthern] Just to prove we're not dead yet

[Ryan Abbenhuys]

The IP allocation system has to be completely overhauled.  I kept saying
this....and saying this...and...

If you want GHO to be a backbone node then yes...it has to be backbone node
in area 0.0.0.0

The problem with nodes being split is where you require sub regions within
regions.  Or you must make area 0.0.0.0 snake it's way carefully through
each region.  The problem with this is that the backbone may require moving
regularley depending on how network topology changes.

What you want to achieve with your decision is the cutting down of
unnecesary route updates.  The routing updates preferably being kept locally
(within clusters where all of the ad-hoc linking is occuring), then what you
have the backbone nodes at the edges of each cluster or region pumping out
are static routes like 10.10.128-129.x is that way---->   and 10.10.40-41.x
is this way  <----.  So adding routes that point towards entire
regions/clusters rather than every individual node in that area.

This requires those running the backbone nodes to maintain the static routes
they are injecting to other areas.  Yes it's manual routing to a degree but
lets be honest, you're at least another 10 years away from having the number
of nodes to require a completely autonomous system if it even gets that far.
At which point if you want you can setup BGP between clusters/regions as by
that stage they are hopefully easier to distinguish and larger.

Sorry if i'm rather blunt...I still hold a grudge against WGrouting claiming
/32's work.


----- Original Message ----- 
From: "Dan Flett" <conhoolio at hotmail.com>
To: "'Ryan Abbenhuys'" <sneeze at alphalink.com.au>;
<rgsouthern at wireless.org.au>; "'Melb Wireless'"
<melbwireless at melbournewireless.org.au>
Sent: Thursday, January 27, 2005 6:28 PM
Subject: RE: [RGSouthern] Just to prove we're not dead yet


> Ryan wrote:
>
> > GHO, and the uplink nodes must be area 0.0.0.0.  Their interfaces
> linking
> > back to the cluster are on the appropriate region 0.0.0.9.
>
> By this do you mean that all nodes that connect to GHO must specify
> their GHO-linked interfaces as Area 0.0.0.0?  This means that any node
> that connects to GHO is a backbone node.
>
> > The uplink nodes will aggregate the routing tables for each region.
>
> I've a question here: GHO's own address block (10.10.129.0/28) is
> allocated from the OEF/Area 0.0.0.9 address space (10.10.128.0/20).
> Won't aggregation of Area 0.0.0.9's routes fail because GHO's addresses
> are smack bang in the middle of OEF's address space?  I thought OSPF
> worked better with route aggregation if each area, including area
> 0.0.0.0, had addresses chosen from separate address spaces.  Shouldn't
> GHO and any other Area 0 interface have addresses chosen from outside
> existing Region Group address spaces?
>
> When I get a chance I'll draw up some diagrams of what I'm talking
> about.
>
> Another issue with OSPF currently:  Johnno's nodes FOA, FOB and FOC are
> in RGSouthern and are linked together.  In a separate cluster we have
> Nodes GMR, GES, GEZ and AFH, also in RGSouthern.  In theory we're all
> supposed to be Area 0.0.0.10.  We're currently linked via another area -
> Area 0 (I assume Node ILE is Area 0).  Should we be virtual-linked?
> OSPF wants areas to be contiguous.  If they aren't there needs to be
> 'virtual-link' or 'shortcut' (not sure which applies best here), and the
> area that is doing the 'virtual-link' or 'shortcut' needs to be
> specified in the config files of the concerned routers.
>
> Also, what area is Node ILE in?  It has one link to OEF (Node GUR) and
> two links to RGSouthern (Node FOB, Node AFH), and no direct links to
> Area 0.0.0.0 (Node GHO).  However, if Johnno made FOB's interfaces to
> GHO and ILE both Area 0.0.0.0 then at least the backbone would be
> contiguous.
>
> This link talks about shortcut ABRs
> http://ietfreport.isoc.org/old-ids/draft-ietf-ospf-shortcut-abr-02.txt
>
> This is why I don't like using OSPF with non-zero areas.  It's simply
> too complicated for us non-network-engineer types.  OSPF expects areas
> to be connected in a certain way.  Nodes in our network often don't link
> up in a way that meets the expected topology - as shown in the examples
> above.  We have to specify exceptions in our config files, and there's
> no reliable way for information about config changes to get to other
> node-owners.  LocFinder doesn't currently support the idea that a node
> can be in more than one area at once, nor does it contain any
> information as to a node's current OSPF configuration.  I suppose people
> could try to keep their up-to-the-minute Quagga config files posted to
> the Wiki, but how can we rely on this?  Again, it seems a bit too hard
> to me.  And it's far too complex a system for a newbie to want to get
> involved in.  A dynamic routing protocol should, at the very least, make
> large-scale routing easier than using static routes.  Using OSPF the way
> we use it seems to make it harder, not easier.
>
> > I have spoken to several cisco nutbags at dimension data who say OSPF
> is
> > fine.  They have said that we could *possibly* use BGP for
> inter-region
> > links at a later stage if need be.
> >
> > pffffffft.
>
> Cool, well it's good we've got some nutbugs on side.  Can you tell us
> what factors they base their decision on that OSPF is fine?  Do they
> realise that 90% of the node-owners don't know much about routing, let
> alone dynamic routing under Linux?  Do they realise that in our network
> OSPF areas can and do become fractured as links appear and disappear?
> Do they realise that there is no single administrative body that could
> control of all routers on the network?
>
> I've outlined my arguments against all-pervasive OSPF in this post and
> in previous posts.  Do they have any counter-arguments?  I don't for one
> second pretend that I have much experience at all to back up my
> arguments.  I only know what I know from reading manuals, articles and
> mailing lists.  The only practical experience I have is playing with
> Quagga on the RGSouthern nodes.  But it seems that already, with the
> recent re-linking of RGSouthern and RG-OEF, that OSPF is becoming
> unwieldy.  I really would like to see some more informed debate on this
> issue.
>
> Cheers,
>
> Dan
> To unsubscribe: send mail to majordomo at wireless.org.au
> with "unsubscribe rgsouthern" in the body of the message
>
>


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