cancel
Showing results for 
Search instead for 
Did you mean: 
cancel
Announcements
Join Customer Connection to register!
1227
Views
0
Helpful
8
Replies
vdadlaney
Beginner

Suboptimal Routing when using a Dual Carrier MPLS WAN

Hi All,

I was reading through the doc http://www.cisco.com/en/US/docs/solutions/Enterprise/WAN_and_MAN/L3VPNCon.html#wp401960 and in this section under dual carrier it says that in a situation where routes are getting redistributed from the PE-CE Routing Protocol (eBGP) into the Local IGP (assume EIGRP) than CE1 and CE2 will learn the route via the local IGP. I am a bit confused when thinking about this because I assume that the eBGP route learnt from the local connection to the SP on that CE would be preferred over the IGP route learnt from the alternate CE since the admin distance of eBGP is better and hence there will never be any suboptimal routing. This behaviour is actually very important if your main sites are acting  as a transit for failure scenarios and I think the desired behaviour to  achieve redundancy. There is one corner case where after a failure due to the weight attribute the eBGP route would not be installed into the RIB however if the weight is set so that all routes from the eBGP neighbor have a higher weight than that case also does not apply.  Can someone please confirm if I am missing something over here. Thx for your help.

Regards,

Vikram

Message was edited by: vdadlaney

8 REPLIES 8
Mahesh Gohil
Rising star

Hello Vikram,

What i understood from link you shared is

                          --------CE3-Site2--CE4--------

CE5--Site1----SP1                                       SP2

                         --------CE1--Site3---CE2------

In above topology Some routes are received by CE1 via. SP1 and CE2 via. SP2 which is advertised from Site2 (CE3 and CE4)

Now Once it is redistributed to IGP the EBGP information will be lost and assume CE1 announce same route back to SP1 which is learned

from CE2. As a result CE5 choose this route via. Site3---SP2---CE4---Site2. Ideally it should move directly to Site2 via. SP1--CE3 because this

route belongs to Site2. This is the case where it takes suboptimal path

Hope it is clear now

Regards

Mahesh

Hi Mahesh,

Thanks for responding. If I understand your explanation correctly that is precisely what I am not understanding and am thinking is not possible because the only way CE1 is going to advertise the route that it learns from its IGP with CE2 is if the eBGP route that it is receiving from SP1 disappears because as long as it has that eBGP route from SP1 it will never install the IGP learnt route in the Routing table. Now I understand that in a failure say Site 2 (CE3) loses its connection to SP1 than CE1 will install the IGP learnt route via CE2 and will advertise that route back to CE5 via SP1 but that is the desired behaviour because Site 1 is only single homed to SP1 and because Site 2 has lost its direct connection to SP1 the traffic from Site 1 to Site 2 needs to traverse via Site 3. Pls let me know if I am making any sense. Thx

Hello Vikram,

am thinking is not possible because the only way CE1 is going to advertise the route that it learns from its IGP with CE2 is if the eBGP route that it is receiving from SP1 disappears because as long as it has that eBGP route from SP1 it will never install the IGP learnt route in the Routing table

Above one is OK if ou

have only single router peering to two different Provider. In the topology however it is three router

(CE1--CE--CE2). CE has visibility of only IGP route and not ebgp route. It will have both route advt. from CE1 and CE2 which is redistributed

from ebgp.  Now CE has two route and b y chance it is having best route via. CE2 which is again advt. to CE1 via. redistribution from igp to ebgp.

Regards

Mahesh

Hi Mahesh,

mahesh.gohil wrote:


It will have both route advt. from CE1 and CE2 which is redistributed

from ebgp.  Now CE has two route and b y chance it is having best route via. CE2 which is again advt. to CE1 via. redistribution from igp to ebgp.

Regards

Mahesh

That would be the case even normally. In order to ensure that your internal CE chooses the route you want it to you have to set the metric in your IGP upon redistribution. I still don't see how the metric on a internal C router in Site 3 affect Site 1 and cause it route via Site 3. The scenario in the URL stated that CE1 in Site 3 which connects to SP1 would advertise the IGP route to Site 1 and in order for this to happen that CE1 router needs to have that IGP route in its routing table which is what I am questioning in that URL because the eBGP learnt route will still be there hence it will never install the IGP route in that CE1 device. Thx for your help

Regards,

Vikram

Hello Vikram,

I haven't found any document which talks about timers of selecting routes from two different protocol.(After how much time router will compare route learned by different protocol) This is what i have seen practically.

Once IGP route is selected against ebgp by router it will advt. to ebgp router and the remote router will not advt. to that peer.

Like PE1---CE1--CE--CE2

Please find summary for above topology

> CE2 Learn route via. ebgp

> CE2 redistribute it to IGP and advt. to CE

> Assume bgp is not yet converged between PE1-CE1

> CE1 have only one route via. CE2

> CE1 advt. that route PE1

> And once PE1 see that route via. CE1 which is best.it will never advt. route to CE1 (seen practicaly)..So no question of comparing route IGP and EBGP

   because PE1 is not going to advt. that route back to CE1

I hope this is clear now

Regards

Mahesh

Hi Mahesh,

Thx for responding back. I am really trying to understand this scenario but I am still having a hard time. Pls see below

mahesh.gohil wrote:

Like PE1---CE1--CE--CE2


> Assume bgp is not yet converged between PE1-CE1

> CE1 have only one route via. CE2

> CE1 advt. that route PE1


I understand the above 2 statements however after that is what I am not sure about


> Assume bgp is not yet converged between PE1-CE1



> And once PE1 see that route via. CE1 which is best.it will never advt. route to CE1 (seen practicaly)..So no question of comparing route IGP and EBGP

   because PE1 is not going to advt. that route back to CE1


You do mention that BGP has not yet converged between PE1 and CE1 so that is the first anomaly that if BGP is not yet converged how would PE1 see the route from CE1 that CE1 learnt from CE2. But that is a afterthought.

Would it be correct to say that CE1 has the route installed as a IGP Route and let's assume that the IGP is EIGRP?

If so than the route is installed in CE1 with a admin distance of 170 (External EIGRP)  and is than getting redistributed into BGP

Once the eBGP neighbor between CE1 and PE1 does come up won't they exchange their routes (this is assuming that PE1 is learning the route from its neighbor PE router that is directly connected to the site that owns that route and not a failure scenario)

Hence won't CE1 see that the route from PE1 will be preferred first because there is a configuration on CE1 that all routes from PE1 have a weight higher than 32768 to account for the corner case and second because its an eBGP route. Hence once CE1 sees that it prefers the eBGP learnt route it will override the EIGRP learnt route due to a better admin distance in the routing table. Thx for your help

Regards,

Vikram

Hello Vikram,

Now we are in sync.

I have performed below test and now agree with what you say.

R3---ebgp---R1----eigrp---R2

> I have announced route 90.90.90.0/24 from R2 via. eigrp

> On R1 it is best because it is having only single route

> On R3 it is best again because it has only one route

> From R3 i announced same route and see below debug

 

      *Mar  1 00:18:39.219: BGP(0): 10.10.10.1 send UPDATE (format) 90.90.90.0/24, next 10.10.10.2, metric 0, path Local <<<< 10.10.10.2 is R1-R3 wan Interface

> On R1 see debug message

    *Mar  1 00:17:55.967: RT: add 90.90.90.0/24 via 10.10.10.2, bgp metric [20/0]  <<<< It has prefered route through ebgp and deleted eigrp route.

Now i can say that the problem description on link you shared is matter of guesswork and not a practical scenario

Anyway thanks for sharing good scenario.

Regards

Mahesh

Hi,

IMHO, there are some errors in that document (where's the CE1 in the diagram, e.g.? My understanding is it's connected to PE4 but that's not 100% clear, it's even shown as connected to Net Y on figure 49).

A realistic problem scenario could be:

Let's say the CE4-PE3 line fails.

CE4 is then receives a prefix X withdraw from PE4.

Consequently, CE1 starts to redistribute the prefix X it received via IGP to BGP.

PE4 advertizes that prefix to SP1 cloud. CE5 receives it.

And when CE4-PE3 line get's Up again, it can happen the prefix advertised from  CE1 has better BGP attributes then the same prefix advertised from CE4 in some part of SP1 cloud (including CE5).

What's the solution?

A careful BGP/IGP redistribution!

You can use tags when redistributing BGP prefixes to the IGP, e.g.

And route-maps prepending an AS number several times (making the AS-PATH long) and decreasing Local Preference and Weight when redistributing those tagged IGP prefixes (only!) to BGP.

This way, you make the prefix X advertised by CE1 less preferred compared to the same prefix advertised by CE4.

And finally, when the CE4-PE3 line get's Up again, CE1 will receive the prefix X via eBGP and will stop redistributing it from IGP.

To make all of this working is a little complicated and requires a good understanding of the BGP Best Path Selection Algorithm (see

http://www.cisco.com/image/gif/paws/13753/25.pdf )

Or, as the article suggests, you can "let SP1 configure local preference or to manipulate any other BGP metrics/attributes so that PE5 prefers the path via PE3 as primary, and reverts to PE4 only if the primary path is lost."

HTH,

Milan