ASR 920/903 Series Routers | Load Balancing

[TrivialPants]

Hi All, 

I am working with TAC on a case currently and we are having trouble getting l2 mpls connections to load balance across a LAG interface. 

Has anyone got any experience with this platform ASR920/RSP1-3 ASR903 devices? I am using the highest license level (advanced metro IP) and I have gone through each type of port-channel load-balance option but with no success. It just seems that we are seeing all the traffic on one link of the 2 10G members. There is some input traffic that is on both (~34Mbps or so) however the outbound traffic is one-sided and only one of the 10G members with ~0.5Mbps on one link and ~2-5Gbps on average on the other link instead of being shared between them.

On the ASR 920 I do not seem to have the option to complete the entropy label commands and assign an actual label number to use entropy labels to better load-balance. 

Load balancing also seems to work just fine on non-MPLS connections.

Does anyone else have this problem, or has anyone run into the same scenario and come up with a solution using these platforms?

[Georg Pauwen]

Hello,

what is your CEF configured as ?

You might want to try the different CEF options available:

ASR920(config)#ip cef load-sharing algorithm ?

include-ports Algorithm that includes layer 4 ports
original Original algorithm
tunnel Algorithm for use in tunnel only environments
universal Algorithm for use in most environments

[Giuseppe Larosa]

Hello @TrivialPants ,

traffic from L2 MPLS services is load balanced based on an hash of the IGP label and the service pseudowiire label.

As a result of this it is normal that a single pseudowire is treated like a single flow and it will use only one link per direction.

I have seen this on ME 3600, ASR 9001, ASR 9006 so I could say that what you see is not a specific limiitation of ASR920  /AR903 platforms but the results of Cisco implementation of load balancing of L2 MPLS services.

To be noted MPLS L3 VPN are load balanced in a fair way because they inspect the inner source IP / destination IP in the payload.

I am surprised that Cisco TAC has not provided you this answer.

To be noted even with ECMP using two L3 links with MPLS over them,  the result would be the same.

Possible workaround is to have multiple pseudowires carrying each a subset of VLANs ,  but this may be feasible or not in your scenario,

Hope to help

Giuseppe

[TrivialPants]

Hi @Giuseppe Larosa - sorry to revisit this so late. I had another instance of this occur, but on a different device, also a Cisco ASR 903 RSP3. However it is using a port-channel to port-channel (20G) to a handoff switch which hosts the mpls pseudowires. Then, there is a 20G OSPF link that is not load balancing, in the same way as these mpls links are not.

However, on this router, the link between the customer handoff router which contains the mpls service instances is load balanced. Just not the ingress/egress 20Gbps OSPF link.

[Giuseppe Larosa]

Hello @TrivialPants ,

interesting followup:

>> However, on this router, the link between the customer handoff router which contains the mpls service instances is load balanced. Just not the ingress/egress 20Gbps OSPF link.

Again the carrier ethernet access link is not part of MPLS domain, so load balancing happen in "ether-channel way" to customer CPE, it does not happen on backbone facing link but it should if there is a pseudowire for each service instance each of them with a different VC label signalled by LDP.

Adding an additional entropy label could be an option on ASR9000, but I'm not sure entropy label is supported on ASR 903 running IOS XE.

There is an RFC about this entropy label that would became the most internal but its usage is to create diversity within a single pseudowire.

All devices perform load balancing based on the inner label for non L3 VPN traffic for L2 services.

Hope to help

Giuseppe