Does load-balancing occur between labeled and unlabeled paths on Cisco IOS ?

[thomastrupel]

Does someone verifed  Luc De Ghein's assertion below ?

If a prefix is reachable via a mix of labeled and unlabeled (IP) paths, Cisco IOS does not consider the unlabeled paths for load-balancing labeled packets. (page 50, MPLS Fundamentals, Cisco Press)

Luc seems to be wrong :

https://trupel.com/networking/Lists/Billets/Post.aspx?ID=5

https://trupel.com/networking/Lists/Billets/Post.aspx?ID=6

Remarks and references are welcome.

[Masoud Pourshabanian]

Hello,

Luc is not wrong. The direction of test in that link is not correct.

The direction in that test is R1-to-R3 while it must be R3-to-R1 to prove or disprove Luc. The reason is R1 receives unlabeled MPLS packets from server(192.168.100.1).

If  a router receives packets with no label, it load balances across both labeled and unlabeled link.

If a router receives packets with MPLS label, it load balances only across labeled link.

If you change the direction from R3 to R1(pinging 192.168.100.1 from 192.168.200.1 instead), you will see Luc is correct.

Masoud

[thomastrupel]

If  a router receives packets with no label, it load balances across both labeled and unlabeled link.

If a router receives packets with MPLS label, it load balances only across labeled link.

Hello Massoud,

Do you have any references ?

Thanks for your help

[Masoud Pourshabanian]

Sorry for the late response. I did not have access to this page. I am not sure why. I received access-denied several times. 

Actually, my second sentence is the same as your sentence and the same as the link below.

http://www.ciscopress.com/articles/article.asp?p=680824

If a prefix is reachable via a mix of labeled and unlabeled (IP) paths, Cisco IOS does not consider the unlabeled paths for load-balancing labeled packets.

I could not find direct mentioning of my first sentence. You can conclude it intuitively. Or you can test it easily by 4 routers located one after after each other. Two middle routers will be connected by two links and edge routers with have several loopback interfaces to ping with different sources and destinations. If you remove MPLS IP, you will see only link with "MPLS IP" will be used for sending the traffic.

Masoud

[thomastrupel]

Hello Masoud,

I think Luc's assertion is platform dependent.

But as you suggested, observing load-sharing between core and edge routers is not the best point of view to prove Luc's assertion.

To confirm or invalidate Luc's assertion (at least for CSR 1000V with IOS XE version 03.14.01.S, IOS version 15.5(1)S1), I'll try another scenario with four routers in Indian file to observe load-sharing between core routers. I'll update this discussion with new results.

Thanks for your help

Thomas

[thomastrupel]

Hello Masoud,

I made a new scenario : https://trupel.com/networking/Lists/Billets/Post.aspx?ID=6

Same observation as in the previous scenario : load-balancing don't occur.

Thomas

[Masoud Pourshabanian]

Hello,

If a packet has two MPLS labels, it can not be routed to its destination if the path is not a MPLS path so ignoring non MPLS path is completely reasonable. New platform might check the number of labels in packets and also consider POP label (P router before  PE) ;however, It has load on router and contradicts the main purpose of label switching which is fast switching. I still think that assertion is correct unless you can try and show it on a new platform
Masoud

[Giuseppe Larosa]

Hello Thomas,

it may depend on Cisco IOS platforms I have seen MPLS L3 VPN connectivity broken by the fact of having multiple equal cost paths labeled and unlabeled.

The devices were C6500 running IOS 12.2(33)SXH, so my personal opinion is that is better to avoid to have equal cost links with MPLS enabled and with MPLS disabled.

Non MPLS links if any, should have higher costs and to be used for special purposes.

The better approach is to have all links with MPLS enabled, so any kind of issue is solved by design.

Hope to help

Giuseppe

[thomastrupel]

Hello Giuseppe,

My intuition is telling me that Luc's assertion is platform dependent.

But as suggested Masoud below, I'll try another scenario with four routers in Indian file to observe load-sharing between core routers. I'll update this discussion with new results.

Thanks for your help

Thomas

[thomastrupel]

Hello Giuseppe,

I made another scenario to observe load-balancing between labeled and unlabeled path at the core level on a plain IPv4-over-MPLS network :

https://trupel.com/networking/Lists/Billets/Post.aspx?ID=6

As you said, it seems Luc's assertion is platform dependent.

Thomas