🚀 go-pugleaf

I am employed by an organization turning from proprietary systems into the
world of UNIX and TCP/IP. One task is a nationwide multiprotocol network.
And therein lies a problem... 

We have been allocated several class B addresses.  The available addresses
may be allocated in three main ways:

- divide the country into regions and allocate one class B address to each
  region  (several minor variations possible)

- allocate all the class B addresses throughout the country in a layered 
  fashion. This proposal has serious drawbacks using the IGRP routing 
  protocol. 

- allocate one class B address to a backbone network and use a remaining 
  class B addresses as regional addresses (similar to the the first proposal)


The entire network is preferably seen as a single network, hence the IGRP
(and later OSPF) is the preferred routing protocol. 

The topology is similar to  (not complete)

                     ----A1---
                    /         \
		   /           \
		  /             \
                 A2              E2
                 |               |
                 |               |
                 |               |
		 B1--------------D
                 |              / \
                 |             /   \
                 |            /     \
                 B2          /       \
                 \           |        |
                  \          |        |  
                   C1--------C2------E1      

                      

Each letter indicated one class B address, e.g. nodes A1 and A2 is within
the same class B address-space.  The allocation is hence following the 
first proposal with regions.  

Each node is implemented with one or more cisco AGS+.  Not shown is the 
local networks connected to each node. 

Each line indicates a physical transmission resource. 

The main problem is efficient use of the available paths in the event of 
failure. E.g. with the proposed address allocation and the use of IGRP
(and OSPF?) it will not be possible to route traffic from A1 to A2 
following the paths A1-E2-D-B1-A2 as traffic from one class B address-space 
cannot be routed to a station within the same address-space via a different 
class B address. Similar problems exist e.g. with traffic from C1 to C2. 

The physical paths cannot be changed. 

One proposal, assuming regional allocation, is the use of secondary 
addresses. I.e. addresses from the A space is used at node E2, D and B1 to 
allow traffic from A1 to A2 to follow A1-E2-D-B1-A2.  Hence, each regional 
network is "stretched" into the other regions - the result is not much 
different from the layered proposal. 

Any comments on 

  - address allocation

  - use of secondary addresses (impact on efficiency?)

  - choice of routing protocol 

will be much appreciated. 

Knut 
====
knutj@idt.unit.no
ta@odin.nta.no