FAT A-VPLS core load-balance

Sherif Atef Ahmed Ismail · ‎01-08-2013

Dears

I need your assistance please regarding A-VPLS core load-balance

As I understood, A-VPLS inserts a new label "Flow Label" so that this label can be used in core routers to acheive load-balancing as shown in figure 2

Now my questions

1- P router has to use port-channel so it can calculate a hash based on the flow label, correct ?

2- Also I should configure "port-channel load-balance mpls" at P router, correct ?

3- If I am correct, flow label is the 3rd label according to figure 1 and according to command reference for "port-channel load-balance label" only last 2 labels are taken into considersation !?

------------Quote----------

If you select label, these guidelines apply:

• With only one MPLS label, the last MPLS label is used.

• With two or more MPLS labels, the last two labels (up to the fifth label) are used.

---------Unquote------------

Regards,

Sherif Ismail

Sherif Atef Ahmed Ismail · ‎01-14-2013

Any ideas

shreerampardhy · ‎01-24-2013

Hello Sherif,

I am not completely sure of the guidelines that cisco have quoted, but the document which you are referring doesnt seem to be updated to the RFC recomendations. According to the RFC 6790, the Flow label ( or the entropy label ) is no longer placed after the application label ( which is the pseudowire label in your diagram ). This is now placed after the tunnel label ( which is the ldp label or RSVP label ). There should also be an Entropy label indicator ( reserved label value of 7 ) after the tunnel label ( and before the entropy label ) to indicate that the following label is the one that will be used for load balancing.

In the initial drafts of this RFC, the egress had to infer from the BOS bit on the application label whether or not there was an entropy label

All the hashing is calculated at the ingress. The output of the hash is the entropy label. The core routers would just use the label to loadbalance the traffic.

Regards,

Shreeram

Sherif Atef Ahmed Ismail · ‎02-04-2013

Hello Shreeram

Many Thanks .. I just saw your reply now

Below is the URL for the document I was referring to

http://www.cisco.com/en/US/solutions/collateral/ns340/ns517/ns224/ns949/ns304/ns975/product_bulletin_c25-602184.html

Actually I am still confused

If I understood correctly, you are saying that location of label should change from 3rd label to 1 or 2nd label

1) Still P rotuers need to be configured with etherchannel , correct ?

2) Each time Ingress router inserts a different Entropy label according to the result of hash function calculated at ingree router, correct ?

Regards

Sherif Ismail

Many Thanks

shreerampardhy · ‎02-07-2013

Hello Sherif,

The entropy labels will now follow the lsp label ( so the locations would be 2 and 3 ). Location 2 would be used by Entropy label indicator ( reserved value 7 ) and the location 3 would be the actual entropy label used for load balancing.

Ingress would calculate a new label when the traffic streams are between different flows. These can be identified by source destination ip address / mac address, or Transport protocols or any other means that are configured on the routers.

The link that you have provided seem to be indicating that the flow label is the last of the label stack. RFC

6790 being pretty new ( it has come out in Nov 2012 ) I am not quite sure if this has been implemented yet.

Regards,

Shreeram

Sherif Atef Ahmed Ismail · ‎02-14-2013

Hello Shreeram

Still intermediate nodes have to be configured with etherchannel to utilize this feature , correct ?

If I am correct, still the problem that Cisco etherchannel hash funchtion do not consider MPLS 3rd label, and this is the label that should be used for load-balancing !

Thanks

Regards

Sherif Ismail

shreerampardhy · ‎02-15-2013

Hello Sherif,

hum.. I have not implemented this solution using cisco etherchannel, but it comes as a surprise for me that cisco etherchannel function do not consider MPLS 3rd label. This is because the third label should have the bottom of stack bit set to 1 and TTL as 0 indicating that this is the flow label. Whenever a device has ECMP links available, it should try to look out for this label to see how it can do the load balancing. I think if you have already impletemeted this solution, you may want to see the load on each link to see if they are loaded approx. same. If you are still seeing only single link being used, then you may want to raise a TAC case to see what could be wrong.

Just for reference,I have captured the output of the packet with flow label on ALU boxes which are also using the last label as flow label.

Regards,

Shreeram