From: Albert Lu (albert_ccie@xxxxxxxxx)
Date: Tue Nov 13 2001 - 07:55:50 GMT-3
Hello All,
I'm considering ways of designing a network with 7 nodes, each node consists
of 2 routers for redundancy and each router has a WAN link to a central hub
site, so all Router1 for all sites will have a WAN link to a hub site, and
all Router2 for all sites will have a WAN link to another hub site. The fact
that there's 2 hub sites for each link is not really relevant in this
discussion, but more along the lines of background information.
| |
| |
R1----R2
The diagram is very simple, basically like what I described earlier but note
that there is a connection (Fast Ethernet) between the two routers. The
whole network is running OSPF, and there is no exposure of it to the
internet, except for using NAT and such.
There are 2 networks being routed throughout the network, lets say a
15.0.0.0 and a 150.0.0.0 along with various subnet derivatives of the two
major networks. The 15.0.0.0 network has to be sent out through the R2 WAN
link and the 150.0.0.0 network has to be sent out R1 WAN link. If one of the
link was to fail, traffic for that network has to be able to be sent through
the other WAN link.
Static routes are not allowed, as using them won't let the routers learn
routes and would cause too much administrative overhead for the network
admins.
If plain OSPF was configured to route the 15.0.0.0 and 150.0.0.0 networks,
then there would be no control on which WAN link each network goes, it would
just follow the shortest metric. I was considering metric manipulation, to
set the one network as a lower metric through a WAN link so that it would
prefer that route. However, I don't like the implication of that metric
propogated to the rest of the other sites.
I was also thinking of setting the bandwidth of one of the WAN links to be
much higher than the other, but that does not guarantee that only one
network would go through the preferred WAN link, in fact both networks could
prefer that one WAN link.
Then I thought of a solution that could work very well, and yet very
simple - "Administrative distance".
Looking at the diagram, OSPF routing updates for both the 150.0.0.0 and the
15.0.0.0 network would come in on both WAN links. Both routers will learn
the routes and propagate those routes to each other through the Fast
Ethernet connection between each other. The route with the lowest metric
will be added into the routing table. There is redundancy with this design,
if a WAN link goes down there will be the other WAN link that the traffic
could go through. However, there is no control of which network goes through
which WAN link.
Lets say that I wanted traffic to 15.0.0.0 to use R2 WAN link, and 150.0.0.0
to use R1 WAN link. Using the 'distance' command in the OSPF process, I
would need to set the admin distance of the 15.0.0.0 routes coming into R1
through the WAN link to 120, this will let the 15.0.0.0 routes learnt from
R2 to be installed into R1's routing table rather than the routes learnt
from the WAN link. The result will be all traffic going towards 15.0.0.0
will go through R2's WAN link.
The same is done on the router for the 150.0.0.0 network to prefer traffic
to use R1's WAN link. Redundancy is built into the design, since if one of
the links went down and removed that route from the routing table, the
second route that was set with a higher admin distance will be installed.
Traffic will divert to the other WAN link. This is similar to a floating
static route used for ISDN backup in alot of the scenarios that alot of you
might have seen, but used not for static routes but for OSPF routes.
I'm going to think about this abit more, and try it out in a lab scenario. I
would really appreciate your comments and any issues you might spot with
this design.
Thanks
Albert
This archive was generated by hypermail 2.1.4 : Fri Jun 21 2002 - 06:45:14 GMT-3