[MLB-WIRELESS] Proposal for a new Melbourne Wireless routing policy

[Dan Flett]

In the midst of all this "why are we here?" discussion, I hope people have
time to digest and respond to some nitty-gritty future-direction stuff. ;)

Here's my proposal for a new, BGP-based routing policy for our network.

This would replace the current OSPF-Area-based policy.

By coincidence, or convergent-thinking, this proposal resembles a one put
forward by Sydney Wireless here:
http://www.sydneywireless.org/index.php?pagename=BackBone&action=PageInfo

I would like this proposal to be reviewed and critiqued by any interested
persons.

Glossary:

AS: Autonomous System - a network of nodes under common direct
administration, using a single internal routing protocol.

ASN: Autonomous System Number - as used in BGP

BGP: Border Gateway Protocol - an external routing protocol

Cluster: A geographically close grouping of nodes.

IETF: The Internet Engineering Task Force

Link: A connection between two nodes

Node: A point of connection to a network

OSPF: Open Shortest Path First - an internal routing protocol


I propose the following:

* Keep the current geographic area-based IP allocation scheme as
administered by LocFinder. (See Note 1)

* Allow any "Operational" or "Testing" nodes to request a BGP ASN from
LocFinder.  LocFinder will keep track of which ASNs have been allocated in
an ASN allocation database, just as it currently does with IP addresses.
ASNs will be allocated from the IETF ASN Private Space - being #64512 -
#65535.  This allows for 1023 BGP "areas". (See Note 2)

* To allow for the possible depletion of private ASN space, nodes within a
closely linked cluster can choose to share a BGP ASN amongst themselves and
use an internal routing protocol such as OSPF.  From a LocFinder point of
view, any node should be allowed to choose the ASN of any of it's link
partners instead of requesting a new, unique ASN for itself. (See Note 3)

* Any node with a single link (a leaf or client node) to one other node does
not need to run any routing protocol at all.  A default route, installed by
the "other" node via DHCP will be sufficient.

* A node with two or more links to nodes within a singe AS cluster does not
need to run BGP, but does need to run the internal routing protocol of that
cluster (e.g. OSPF).

* A node which makes links to nodes outside its home AS cluster needs to run
BGP.  The node configuration will need to know the following:
 ** The nodes' own ASN
 ** The ASNs of it's link partners
 ** The IP addresses of other BGP speakers in the nodes' home AS - this is
so that the node can use iBGP to coordinate eBGP routing information to
other border gateways.
 ** Internal routing protocol configuration and redistribution of routes
between BGP and the Internal Routing Protocol (e.g. OSPF)
 (See Note 4)

Notes:

1. I no longer believe that the region group supernetting used in the
current IP allocation scheme will help with routing by allowing the
aggregation of routes.  The nature of wireless networking means that it is
impossible to predict where clusters will form and which areas will link to
which.  I can see no easy way of allocating IP addresses to enable route
aggregation.  However I believe that we can just accept this as a fact of
life.  The current IP allocation system isn't broken and doesn't need to be
changed.  If in the future the address space of any area becomes depleted,
the unused space of other areas could be used.

2. I believe it will be a long time before we have more than 1023 routing
nodes on the network.  So until that happens, do we need to run internal
routing protocols at all?  Will a BGP-only network work, where each routing
node has a unique ASN?  By the time we reach 1023 routing nodes, the BGP
protocol will likely have been updated to allow more ASNs overall.

3. Clusters that use OSPF should enable simple authentication on their
interfaces - and the authentication password should be that cluster's BGP
ASN.  This will prevent OSPF in neighboring clusters from communicating
directly, and will help prevent OSPF from being destabilized by
misconfigured or default-configured routers.  I strongly recommend NOT using
non-backbone OSPF areas as these add an extra layer of administrative
complexity.  The only OSPF Area in use in any cluster should be Area 0.

4.  As suggested in the SydneyWireless link above, LocFinder could calculate
and supply a nodes' configuration information in an XML file, in
human-readable format.  The node-owner could either enter this information
manually, or software running on the node could parse the XML file and
configure itself.

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