EIGRP slowness and backup problem

[jmarkotic]

Hi,

customer has bunch of location connected to regional office. Regional office nicely summarize all routes for a central location. On central location there is

dialin router that handles location when they lose primary link and connect via isdn (centralized backup).

When location goes on backup, it sends its eigrp route so that evreybodu knows it is on backup.

So on central routers there is summarized route from regional office and more specific one from remote location on backup.

Traffic for that location is routed ok thru backup.

When primary link of that location goes up, location starts to send packets thru

primary link and isdn timer start to decrease. As isdn link is still up (because of delay timer) specific eigrp route is still sent to a central location so during a transition time

we have asymetric routing - location sends packet trhu primary link, and packets are returned thru isdn dialin router because of more specific route.

Eventually location shuts down isdn interface.

But although isdn connection is down, eigrp neighborship between dialin router and

location via isdn is not immediately broken (missed eigrp hello packets) so specific route is still in routing table as location is still connected via isdn. So when traffic is returned from

central location to a remote location, packet is routed according to a still living specific route and dialin router on central location calls remote location (they can call each other) and isdn connection is restored.

If there was no traffic during isdn going down and neighborship breaking, route would dissapear and it would be fine. This way customer has ISDN costs higher than it should be because isdn is very often up.

Solution: we could remove summarization done on regional office toward

central location so all specific routes would be visible on central

location. But design studies always recommend summarization in direction to

central location. I know there are lot of wasy to avoid this but I am open to suggestions.

thanks,

[oleg_vk]

Hi there,

What I would suggest:

1. Move dial backup point to regional offices. It may reduce cost of calls but increase cost of additional router/card-in-router for RO.

2. Move summarization point to central office. If you are not propagating full rouitng table from CO to RO/BO, this would not introduce additional load on RO?BO routers. And I beleive you have good enough routers in CO to handle all routing table.

3. "Play" with EIGRP timers on dial-backup router. This will not eliminate your problem completely, but you can reduce "downtime".

[jmarkotic]

Hi Oleg,

thanks for good suggestions.

1. Moving dial backups to RO would mean I would need another dedicated dialin router for backup calls. If it would be one router, than crasching one router would leave offices withot backup also. Also, dial backup costs are only marginally higher.

2. This is exactly what I did. I am filtering and sending only default route to RO/BO.

3. Tried that (eigrp hello/dead) and I definitely lowered possibility for problem, but problem is still the same if continuous traffic goes during failover.

It already works this way. But right now when some location lose connection, eigrp queries are sent thru core, and if I had summarization on RO ... but nevermind.

thanks again,

j