cancel
Showing results for 
Search instead for 
Did you mean: 
cancel
278
Views
25
Helpful
9
Replies
Highlighted
Beginner

BGP Link Bandwidth - Problems (or not?) to load balancing.

Guys, i'm recently studying for an implementation of BGP Link Bandwidth so that i can be able to load balancing traffic between two point to point networks that i have on my company.

What i have reached so far is that I'm able to see the networks via sh ip bgp x.y.w.z and i can see two paths with equal traffic share count.

The problem is that i can see the traffic being load balancing between the two paths:

1. Extended ping between remote networks are only using a single path.

2. "sh ip route x.y.w.z" is only presenting a single route for that traffic.

 

Example:

 

R1#sh ip bgp 10.40.1.1
BGP routing table entry for 10.40.1.0/24, version 3342
Paths: (2 available, best #2, table default)
Multipath: eBGP iBGP
Advertised to update-groups:
12
Refresh Epoch 1
64828
172.19.240.41 (metric 12) from 172.19.240.41 (172.19.240.41)
Origin incomplete, metric 11, localpref 100, valid, internal
DMZ-Link Bw 18125 kbytes
rx pathid: 0, tx pathid: 0
Refresh Epoch 1
64828
172.19.240.185 from 172.19.240.185 (172.19.240.252)
Origin incomplete, metric 11, localpref 100, valid, external, best
DMZ-Link Bw 18125 kbytes
rx pathid: 0, tx pathid: 0x0

 

But "sh ip route" presents only a single route:

 

R1#sh ip route 10.40.1.1
Routing entry for 10.40.1.0/24
Known via "bgp 64864", distance 20, metric 11
Tag 64828, type external
Redistributing via ospf 100
Advertised by ospf 100 subnets
Last update from 172.19.240.185 00:10:34 ago
Routing Descriptor Blocks:
* 172.19.240.185, from 172.19.240.185, 00:10:34 ago
Route metric is 11, traffic share count is 1
AS Hops 1
Route tag 64828
MPLS label: none

 

My diagram is in annex. but what i do have is a L3 switch advertising via OSPF to two routers that in time are injecting the OSPF routes into BGP. The other networks follows the same schema.

 

What i think it's possible the problem is that maybe BGP Link Bandwidth only works with 3 routers running IBGP so that the router that isn't connected to directly to the eBGP is the one that really load balancing the traffic.

 

In annex you can find all the running configs and diagram of the implementation.

 

Would love to have some help and thanks in advance.

9 REPLIES 9
Highlighted
VIP Expert

Paths: (2 available, best #2, table default)

 

Origin incomplete, metric 11, localpref 100, valid, internal

 

Origin incomplete, metric 11, localpref 100, valid, external, best

 

I recall, with BGP, external is consider better than internal.  I.e. those two routes are not considered equal.

Highlighted

but is there a way to make they at least think that they are equal?

Highlighted

Yes, the easiest way (as also suggested by Paul) would be to do the ECMP from R5 and R7. As your diagram shows a non-public BGP network, the size of the route table should allow it to be passed inward.

BTW, in past I've deal with a similar situation, for Internet edge routers, accepting full Internet tables (from different providers), and not desiring to pass that route table further toward the inside, just by having the next hop interior router have a default route to both EBGP routers. If one of providers had a shorter AS path, that traffic would go laterally from one EBGP router to the other.
Highlighted
VIP Mentor

Hello
I think you want to LB (equal/unequal) from each internal ibgp rtr not on the ibgp/ebgp rtrs.
Please see attach file for possible changes to your existing  configuration.



kind regards
Paul

Please rate and mark posts accordingly if you have found any of the information provided useful.
It will hopefully assist others with similar issues in the future
Highlighted

Thanks Paul.

I would like to do that but unfortunately i dont have BGP running on the switches. just ospf.

Highlighted

Hello

Apologies i was just looking at youe topology and assumed the R5/R7 were ibgp rtrs



kind regards
Paul

Please rate and mark posts accordingly if you have found any of the information provided useful.
It will hopefully assist others with similar issues in the future
Highlighted

Do you have OSPF running on R1..R4? If so, and if the BGP table isn't large, you could redistribute your BGP routes into it. If the BGP table is large, perhaps you could summarize or perhaps even inject a default route from those routers into OSPF.

If you don't want the OSPF routes to go beyond R5 and R6, you could run a separate OSPF process (or VRF) and redistribute the BGP routes (i.e. still using OSPF) just to it, so only R5 and R6 "see" R1..R4 routes.
Highlighted

Thanks Joseph.

In fact what i'm doing is something similar.

I'm running an internal OSPF process on the routers that is responsible to redistribute the routes do the BGP and vice versa.

Problem is that when i send traffic to one of the routers this router doesnt balance the traffic with the other. Much likely because the preference of the external BGP as already mentioned - i think by you.

Highlighted

Correct, the problem is you cannot easily balance traffic on R1..R4. However, if you're injecting your BGP routes, into OSPF, seen by R5 and R6, assuming you're injecting with the same OSPF metrics, R5 and R6 should see all the routes as ECMP to their next hop, R1..R4. That's not the case?
Content for Community-Ad