MPLS FAT Label


Userlevel 2
Hello,

By reading this article http://packetpushers.net/fat-or-entropy-label/ I've been wondering if Extreme could implement something like that (FAT Label)?

Some l2vpn circuit are getting 10Gbps and the outer interfaces port channels aren't handling it well.

Thanks 🙂

12 replies

Userlevel 7
Hi Julian,

have looked at the GTAC Knowledge article Incorrect load sharing on MPLS P-node?

Erik
Userlevel 6
There is another GTAC article that might help:

https://gtacknowledge.extremenetworks.com/articles/Q_A/Why-the-traffic-is-not-getting-load-balanced-...
Userlevel 2
So, we've tried everything is possible to solve that, but none solved.
As the sharing only see one label carrying that huge traffic, it cannot load-balacing between the aggregates ports.
I think the only way to do that is by adding this on the mpls header (Entropy/Fat label).
https://tools.ietf.org/html/rfc6790
Userlevel 3
Julian Eble wrote:

So, we've tried everything is possible to solve that, but none solved.
As the sharing only see one label carrying that huge traffic, it cannot load-balacing between the aggregates ports.
I think the only way to do that is by adding this on the mpls header (Entropy/Fat label).
https://tools.ietf.org/html/rfc6790

Hi Julian,
did you try to use "l2vpn sharing" feature?
Please check examples from discussion on a link below. It might help you to solve the problem.
https://community.extremenetworks.com/extreme/topics/load-balance-problem-using-vpls-as-transport

Keep in mind that that feature has some limitations:
https://gtacknowledge.extremenetworks.com/articles/Q_A/What-are-limitations-for-Pseudowire-Label-Swi...
Userlevel 2
Julian Eble wrote:

So, we've tried everything is possible to solve that, but none solved.
As the sharing only see one label carrying that huge traffic, it cannot load-balacing between the aggregates ports.
I think the only way to do that is by adding this on the mpls header (Entropy/Fat label).
https://tools.ietf.org/html/rfc6790

Hello Konstantin,

It's odd because I did the same thing in my lab and couldn't get the same result. What version did you use in that test enviroment?
Another thing, is that I'm waiting for xos0066476 to be solved, isn't save to use rsvp in the current working network.
Userlevel 3
Julian Eble wrote:

So, we've tried everything is possible to solve that, but none solved.
As the sharing only see one label carrying that huge traffic, it cannot load-balacing between the aggregates ports.
I think the only way to do that is by adding this on the mpls header (Entropy/Fat label).
https://tools.ietf.org/html/rfc6790

When I started to test it I also failed and then I figured out that LSP sharing doesn't work with unknown unicast traffic:
https://gtacknowledge.extremenetworks.com/articles/Solution/Pseudowire-Label-Switch-Path-Load-Sharin...

So make sure that traffic which you trying to load-balance is bidirectional.
Userlevel 2
Julian Eble wrote:

So, we've tried everything is possible to solve that, but none solved.
As the sharing only see one label carrying that huge traffic, it cannot load-balacing between the aggregates ports.
I think the only way to do that is by adding this on the mpls header (Entropy/Fat label).
https://tools.ietf.org/html/rfc6790

Most of our traffic is one source/dest mac but has multiples source/dest IP unicast
some of these vpls has 15Gb.
Is there any problem of what kind of unicast traffic is? even if it's just one mac and lsp will do the balancing?
I'll give a new try to that in my lab..
Userlevel 6
Could you please share your network diagram and point where the balancing is not working as expected?

This will help us to have a better insight.

Thanks.
Userlevel 2
Henrique wrote:

Could you please share your network diagram and point where the balancing is not working as expected?

This will help us to have a better insight.

Thanks.

Henrique,

The backbone isn't load balancing.

A ticket was answered by you informing that behaviour(Unifique).


Houve a resposta informando a impossibilidade e referente ao uso de dois labels.
Queria apenas discutir a possibilidade e se isso poderia ser feito da parte de entropia e adição de label para o hash.
Userlevel 6
Henrique wrote:

Could you please share your network diagram and point where the balancing is not working as expected?

This will help us to have a better insight.

Thanks.

Hi Julian, I'm not sure if there is a roadmap to have "The Use of Entropy Labels in MPLS Forwarding" implemented.

I will check internally and let you know.
Userlevel 6
Henrique wrote:

Could you please share your network diagram and point where the balancing is not working as expected?

This will help us to have a better insight.

Thanks.

Hi Julian, currently there is no roadmap to have that feature implemented. I will let you know fi anything changes during the next months.
Userlevel 2
Hi Henrique.

Can you let me know whether would it be possible to do using x460(not x460-g2)?
I have a core with more than 4 of them, and I couldn't do that, I've been lost in find out some solution for my Extreme Network, otherwise I would have to replace those one.

tks

Reply