We have 4 unattended locations connected over leased lines (ring topology). From two other locations we have remote-access over a leased line and ISDN for backup to the 4 nodes. Network layout:
C ----- A ---(leased line)--- Mngt_1
D ----- B ---(leased line)--- Mngt_2
The remote-access from Mngt_1 and 2 over the leased line should be used to reach all 4 nodes. When this leased line failes, we have an ISDN backup. Further more we also have the possibilty to reach all nodes directly over ISDN (in case the backbone between A-B-C-D fails and the nodes are not all reachable over the normal remote-access connection).
On the backbone routers A-D I use EIGRP. On the two Mngt routers I use static routes and dialer interfaces to the 4 nodes (one dialer is configured as isdn backup for the leased line). On the 4 nodes we have dedicated routers Cisco 2610 for the remote-access, also configured with dialer interfaces. My problem is now, that when I access e.g. node D from Mngt_1 over the leased line, node D best route back to Mngt_1 is over the ISDN connection and not over backbone to node A and then to Mngt_1. How can I solve this problem, that the ISDN connections are only used, when the backbone fails (e.g. node D is isolated)?
Any help is very welcome.
I would use a floating static rather than ISDN backup lines.
Set up the ISDN lines with IP addresses like you were doing a remote access connection.
Include the management stations in the EIGRP, and let the EIGRP take care of the routing. Then make a static route back from router D to Mngt_1, but using an administrative distance more than 90.
The EIGRP route will be in the routing table with AD 90. When the route fails, the static route will replace it with an AD 180.
Thank's for your answer.
Sorry, I forgot to mention that I've already tested it the way you described. I used EIGRP over the leased lines to the management locations and configured floating static routes for the ISDN connections. This works basically fine for the normal case. But if one of the leased lines to the management locations fails, I got "strange" routing behaviours. E.g. if the leased line to Mngt_1 failed, an ISDN connection to node A was established, but node A had a routing entry over node B and Mngt_2 to reach Mngt_1 (because EIGRP routes from Mngt_2 have a metric of 90 and are still better than the floating static routes).
That's why I'm confused now, how to solve this problem.
Do I take it that Mngt_1 and Mngt_2 are in the same network then? I'm not sure I fully understand what happened in your floating-static experiment. If node A still had an EIGRP route to Mngt_1 via B and Mngt_2, and presumably Mngt_1 had a route to A via Mngt_2 and B, then I don't really understand why the ISDN came up at all. Normally it would not be dialed until it is actually required for routing a packet, and the packets were still being routed through the EIGRP routes.
Unless of course you had the backup command still on the dialer?
This might be off the hook a little bit, but here it goes anyway. I agree with Kevin that it seems strange that there is a route to the other mgmt station from Mgnt_2, so there must be a link between the two. Anyway, I would think EIGRP is not converging fast enough to delete that route before the ISDN line comes back up, if that is the case you might want to use the backup delay command to force the ISDN line to come after let's say 30 seconds so all routes are on the passive state by then.
Mngt_1 and Mngt_2 are different networks and have no direct link to each other. On the routers at Mngt_1+2 I use the isdn backup command and on the serial interface. On the routers in node A-D I've configured a floating static route with metric 200 back to the Mngt-routers. The ISDN backup line comes up, but the routing back to the Mngt location is not working, because the routing table e.g. on the router in node A has a route to Mngt_1 over node B (better metric from the EIGRP learnd foute from router node B than the floating static route on the local router)!
There seems to be a disconnect here... where is the route to Mngt_1 from node B coming from? Does node B have a route (in which case it should be used) or does it not have a route (in which case you have a stuck route which is not getting removed from the routing table, a MAJOR problem which you need to resolve before going any further)?
The scenario you describe SHOULD be easily solved using standard DDR dial backup. So something is not as described.
Vincent C Jones
When EIGRP is configured also on Mngt_1 and Mngt_2,
the route to Mngt_1 is distributed from the router at Mngt_2 to all nodes A-D. Beside this EIGRP route entry, all nodes are configured with floating static routes to Mngt_1+2, but with a higher metric value.
But what's confusing me is, why is a route from e.g. node A over Mngt_2 better than the direct static route to Mngt_1 (in case the leased line to Mngt_1 failed).
Mngt_2 should only be advertising a route to Mngt_1 if it actually has a working route. In which case, why not use the ISDN link which is already up rather than bringing up yet another ISDN link?
As for your confusion: you need to study the difference between administrative distance, route metric, and prefix length; and how each is treated in populating the forwarding table actually used in a router to route packets out the correct interface.
Enjoy your reading :-)
Vincent C Jones
I think I understand the problem a little bit better now, but one question: Are you redistributing your static routes into EIGRP? That will explain why you are getting the route on router B through Mngt_2.By the way, you are configuring an administrative distance on the static routes, not a metric. EIGRP uses a composite metric that will take into account BW, MTU and delay over your ISDN lines. If you are redistributing your static routes you might have to filter them so you do not get into a routing loop for certain routes.