Parallel BGP sessions with the same peer

jalalyounus · ‎03-26-2012

Hi Everyone;

I hope everybody is keeping well.

I have a situation here.

I have a client hwo is unning a bgp session with us using a 1G ethenet connection (asr9000).

The same client now wishes to have another bgp session with us on a 10G interface (asr 9000).

I had assigned them a temporary IP prefix from a /29 range. The original bgp peering (on 1 g connection) uses a /30 range.

Current configs on my side are as follows:

neighbor 86.107.28.242 (10 G)

remote-as xxxx

use neighbor-group abc

description xxx

address-family ipv4 unicast

route-policy bbb in

route-policy ccc out

neighbor 86.107.23.210 (1G)

remote-as xxxx

use neighbor-group abc

description xxx

address-family ipv4 unicast

route-policy bbb in

route-policy ccc out

There is no intention to implement bgp laod balancing feature (using mltipath or otherwise). Do you think this config would work fine?

regards

Alexei Kiritchenko · ‎03-26-2012

Hi,

Yes, that would work. You’ll get 2 sessions with absolutely identical parameters for all routes except the peer’s IP address. Hence, the best path would be the path with the lowest peer address, routes from 86.107.23.210 would be taken as the best in this case.

BGP Best Path Selection Algorithm

http://www.cisco.com/en/US/tech/tk365/technologies_tech_note09186a0080094431.shtml

Regards,

/A

jalalyounus · ‎03-26-2012

Thanks Alexei.

So I do not need to set multipath value to 2? if not, please advise how the peering would get established?

regards

Alexei Kiritchenko · ‎03-26-2012

Hi Jalal,

Your config from the first post would work just fine. Source and destination IP addresses are different, so there are no any issues making both sessions UP even if physically that is the same router.

Regards,

/A

jalalyounus · ‎03-26-2012

Thanks alot Alexei again.

Just one thing, i wanted check with you. I read somewhere that 2 bgp neighbors connected through physical interfaces(not loop backs), can only have one active bgp peerig session between them. If two same cost links are used to i interconnect them only one would work.

Since in my scenario, there two pysical connections between the two neighbors and these connections are not equal cost links. One is 1G and the other is a 10G. I wonder what might happen here, which link would be chosen by bgp to select the routes to be installed in the routing tbale of the client?

regards

Alexei Kiritchenko · ‎03-26-2012

BGP would choose a path based on BGP Best Path Selection Algorithm

http://www.cisco.com/en/US/tech/tk365/technologies_tech_note09186a0080094431.shtml

in this scenario, the best path would be the path with the lowest peer address, hence routes learned from 86.107.23.210 would be taken as the best (compare to routes from 86.107.28.242 ) in this case and installed into the routing table.

Regard,

/A

jalalyounus · ‎03-26-2012

Thanks again boss.

Would they still be able to see the routes coming in through the 10 G link though?

regards

Alexei Kiritchenko · ‎03-27-2012

What you get is something like this. BGP will learn same network over 2 sessions. Here the net 3.3.3.0/24 was learnt from 100.100.46.6 and 100.100.64.6 (that is the same router here)

Origin codes: i - IGP, e - EGP, ? - incomplete

Network Next Hop Metric LocPrf Weight Path

*> 3.3.3.0/24 100.100.46.6 0 0 10 ?

* 100.100.64.6 0 0 10 ?

100.100.46.6 is taken as the best (marked with > ) as the IP address is lower compare to 100.100.64.6

RP/0/0/CPU0:R4#sh bgp 3.3.3.0/24 bestpath-compare

Tue Mar 27 10:46:12.441 EEST

BGP routing table entry for 3.3.3.0/24

Versions:

Process bRIB/RIB SendTblVer

Speaker 16 16

Last Modified: Mar 26 17:52:21.000 for 16:53:52

Paths: (2 available, best #1)

Advertised to update-groups (with more than one peer):

0.2

Path #1: Received by speaker 0

Advertised to update-groups (with more than one peer):

0.2

10

100.100.46.6 from 100.100.46.6 (192.168.0.6)

Origin incomplete, metric 0, localpref 100, valid, external, best, group-best

Received Path ID 0, Local Path ID 1, version 16

best of AS 10, Overall best

Path #2: Received by speaker 0

Not advertised to any peer

10

100.100.64.6 from 100.100.64.6 (192.168.0.6)

Origin incomplete, metric 0, localpref 100, valid, external

Received Path ID 0, Local Path ID 0, version 0

Higher neighbor address than best path (path #1)

The routing table has only the best path installed

RP/0/0/CPU0:R4#show route 3.3.3.0/24

Tue Mar 27 10:46:47.425 EEST

Routing entry for 3.3.3.0/24

Known via "bgp 65000", distance 20, metric 0

Tag 10, type external

Installed Mar 26 17:52:20.911 for 16:54:26

Routing Descriptor Blocks

100.100.46.6, from 100.100.46.6

Route metric is 0

No advertising protos.

If the 1G link fail, the routing table would install routes pointing to 100.100.64.6.

But you can modify the behavior and distribute the routes among the links as you wish or put a ebgp multipath to load balance between them.

for example here is the routing table when ebgp multipath is enabled:

RP/0/0/CPU0:R4#show route 3.3.3.0/24

Tue Mar 27 11:05:08.512 EEST

Routing entry for 3.3.3.0/24

Known via "bgp 65000", distance 20, metric 0

Tag 10, type external

Installed Mar 27 10:59:56.936 for 00:05:12

Routing Descriptor Blocks

100.100.46.6, from 100.100.46.6, BGP multi path

Route metric is 0

100.100.64.6, from 100.100.64.6, BGP multi path

Route metric is 0

No advertising protos.

Regards,

/A

dipankar · ‎01-16-2018

Hi,

just One query,

in this Case, BGP Prefer routes from the peer with the lowest peer IP address.

here the selection will work ( between 2 IPv4 peer IP's)

in case, I have one BGP neighbor is IPv4 and One more neighbor ship with IPv6.

and getting the same prefix x.x.x.x/24 via both the neighbors , How selection will work !!!

BR//
Dipankar.

nkarpysh · ‎01-16-2018

Hello,

I guess what you mean is that you carry one address family - e.g. IPv4 using IPv4 and IPv6 transport. Situation is that IPv4 NLRI supports only IPv4 NH and IPv6 NRLI supports only IPv6 nexthops. Thus using this approach you will still need to use route-map to modify transport NH to the one understood by your address-family. Meaning for IPv4 AF you will keep IPv4 transport NH but modify IPv6 transport to IPv4.

In this case you will again end up comparing IPv4 against IPv4 and IPv6 against IPv6 as in your first example.

HTH

Niko

HTH,
Niko

dipankar · ‎01-28-2018

hi,

I guess what you mean is that you carry one address family - e.g. IPv4
using IPv4 and IPv6 transport.- Correct

Situation is that IPv4 NLRI supports only IPv4 NH and IPv6 NRLI supports
only IPv6 nexthops. - correct

Thus using this approach you will still need to use route-map to modify
transport NH to the one understood by your address-family. correct

Now my actual Query.

R1(2069::1/64)------(2069::2/64) R2

R1 loopback 1.1.1.1 and 2021::1/128

R2 loopback 2.2.2.2 and 2022::1/128

I have 1 BGP sessions on IPv4 TCP ( from 1.1.1.1 to 2.2.2.2) and AF is IPv6.

I have 1 more BGP session on Ipv6 TCP ( from 2021::1 to 2021::2) and AF is
IPv6.

Now here for a certain Route 2000::/64 is coming from R1 to R2, i will get
2 Peer address

- NH is 2069::1 and Peer is 1.1.1.1 ( changed the NH using map)

- NH is 2021::1 and peer is 2021::1

So how the selection will happen on the basis of Peer address ( i mean how
i will compare a ipv4(1.1.1.1) and IPv6 address(2021::1))