Routing

We are testing out the abilities of the MX960 with some trio MPC's. We currently are running into an issue with QoS and customers that would terminate their L3 services at our MX960. To our knowledge, and everyone we've spoken to at Juniper there's only 2 modes for AE's, replicated and scale mode. Both of which would not allow us to give a customer their exact purchased bandwidth. The gyst of the scenario is, an AE is the transit path for either an IP on the IFL of the AE, or an IRB being on a bridge domain that transits an IFL on the AE. How do we give customers no matter the amount of flows, the exact amount of bandwidth they’ve purchased?

Using 100mb for example and an AE with 4 links

 - In replicated mode they could use up to 400mb of traffic. 1 flow would be limited to 100mb, but multiple flows can consume the 400mb on each link member giving it an aggregate of 400mb possible.

 -In scale mode they could use up to 100mb traffic, but in order to use 100mb they would have to have multiple flows moving across all 4 participating links at 25mbps. And each flow would be limited to 25mbps at the most.

 

Is this really the only two outcomes? Am I incorrect for saying that on Cisco ASR's we can use a bundle-ether and would be able to give them 100mb exactly, regardless the amount of links or flows used?

Yes, these two options are exactly the ones to choose from. When a scheduler-map is applied on a bundle, the scheduler-map is applied at each member interface and there is no synchronization between the member interfaces about the physical interface queues. Only way to achieve your goals would be the scale mode in combination with per-packet load balancing, but Juniper does per flow load-balancing only, so it does not work. By the way, up to my knowledge behaviour on Cisco ASR9k is exactly the same (only supporting per flow load balancing across bundles and those two options).

 

In case you only want to do policing (rate-limiting) instead of shaping across the member links, that would work on the MX series by using the shared-bandwidth-policer option.

 

Cheers,

Carsten

Thanks, we just couldn't get a straight answer and every time we talked to someone it was yes you can, or no you can't do that answers. It is the same story with other vendors after doing some more research. I did find some info on MX router adaptive load balancing which would seem to be what you describe, and then there's per packet spraying load balancing which is considered a last resort and certainly not something we'd want on a voice network.

Hi, There is only workaround... I.e. do the active and standby kind of configuration under ae interface with lacp with 2 member links. Through which you can achieve your objective.

Regards,
Vinay

Correct and thats what we've moved towards.