Redistributing OSPF Metrics

[Patrick McHenry]

Hi,

I'm setting up a 1001 router at a branch office with a DS3 WAN interface that will connect to a 1006 at the HQ. We are starting to migrate to EIGRP in the process so, I'm going to set up the site with EIGRP and redistribute OSPF into the EIGRP and EIGRP into OSPF at the HQ 1006 router. The Branch subnet and the WAN subnet connecting the Branch to the HQ will be EIGRP - the rest of the HQ router will be OSPF

My questions are:

If the connection between the Branch and HQ is a DS3, what should I sent the metrics for when redistributing OSPF into EIGRP at the HQ?

Also, we have started the migration over to EIGRP with our DMVPN environment and we have called it EIGRP 99. Will it be ok to call this new pocket of EIGRP also 99? I understand that for EIGRP to work, the AS must be the same but, these instances are not connected. This won't cause any problems will it?

Thank you, Pat.

[Richard Burts]

Pat

It certainly would not cause any problem to call both sets of EIGRP 99. They may not connect today but what is the possibility that as things go forward they will become connected? If that is possible then it is a real point in favor of using 99 for both now.

You can pick any value for the metric used in redistribution. There is not much point in trying to pick something that matches the speed of the link that you are using. My advice is to pick some fairly high number for the metric. Many people pick very low (attractive) values for redistribution. My viewpoint is that we really do not know what is the true metric for a redistributed route so I want to know that it is available but if there is a route that I really do know its true metric I would rather use that one.

HTH

Rick

[Patrick McHenry]

Richard,

What metrics would you give the redistributed OSPF networks?

Thank you, Pat.

[Richard Burts]

Pat

For the metrics in redistribution I might use something like 15000 20000 1 1 1500. That gives it a pretty low bandwidth, a pretty high delay, and the other values do not matter particularly.

HTH

Rick

[Patrick McHenry]

Richard  - Thank you for the response.

I'm not sure I follow your reasoning behind creating a high metric. Why would I want to make the OSPF routes coming from HQ look unattractive to the Branch router? Could you explain a little?

Thanks, Pat.

[Richard Burts]

Pat

My reasoning is that with a redistributed route the important thing is that your routing logic should know that the destination is reachable through this (redistributed) route. But since we do not know what the "true" metric should be we have to make one up. If we make up a metric with a low number we make this look like a very attractive route, even though we do not know for sure that it is.

Let me suggest an example. Let us assume that you are in New York City and you want to send a data packet to Washington DC. Your routing table has two entries. There is one entry in the routing table for a route that is generated entirely in EIGRP. From that routing metric we can know that the packet will go from New York to Philadelphia and from Philadelphia to Washington - so it has a metric of X. The other entry in the routing table was advertised to you from Newark NJ where it was redistributed from OSPF. Its metric is N (which is lower than X) and looks better. What is not apparent is that the OSPF route from Newark goes through Cleveland Ohio to go to Atlanta Georgia, to go to Washington). Which one would you really want to use?

In more general terms I am suggesting that when we have to make a choice and we have an accurate metric for one choice and the other metric is something made up that it is safer to go with the choice whose value we know is accurate. That is what you get when you make the metric for redistribution high. If you choose a low value for the redistribution metric then you are saying that in the choice you would rather choose the one not really known because perhaps it is better than the one that you know.

HTH

Rick

[Patrick McHenry]

Rick,

Alright - I understand what you were getting at. And as for best practice I think you are right but, at this point, the one WAN interface is the only path to HQ so, I don't think it really matters, correct?

What if down the road we add another WAN link from this branch site and that path is an EIGRP route that is slower than the redistributed OSPF routes that we artificially made unattractive and it takes the EIGRP route because of the more attractive metric?

Thanks, Pat.

[Richard Burts · Accepted solution]

Pat

When there is only one path to choose from then it does not matter whether you set the metric for redistribution high or set it low. Your choice will select the best route. But what about situations where there is more than one choice?

As I attempted to discuss in my previous post I believe that the essence of this question is about a choice if you need to make a decision (a routing decision especially) and you have two choices - one choice is an accurate description of the alternative and the other choice is an unknown, then which do you want to choose?

One thought that comes to my mind is a scenario for a game show on TV. You are the contestant and in front of you are two doors. Behind one door is a prize, whose value is known. Behind the other door is a prize whose value is unknown. Which do you choose?

If you systemically like the possibility that the unknown might be the best then you set the metric for redistribution to be very low - and you will choose the redistributed route when it is available. Sometimes you win and sometimes you do not win. If you prefer to choose the known alternative then you set the metric for redistribution to be high - and you will choose the native route when it is available. sometimes you win and sometimes you do not win.

You can always come up with scenarios where setting the metric for redistribution to a high value is the best choice. And you can always come up with scenarios where setting the metric for redistribution to a low value is the best choice. And in real life sometimes it is one and sometimes it is the other. Sometimes your choice will be the optimum choice. And sometimes it is not. So are you a gambler who will choose the unknown because it might have the best payback or are you conservative who will choose the known (safe) choice? And which way do you want your network to operate?

HTH

Rick