[c-nsp] ASR9k Bundle QoS in 6.0.1

Mark Tinka mark.tinka at seacom.mu
Sat May 14 07:35:53 EDT 2016



On 14/May/16 11:22, Saku Ytti wrote:

> I'd love if you'd have time to ask JTAC their explanation why it
> didn't work. Really curious.

I did speak to JTAC, my SE as well as the MX BU, and they were mostly
dumb-founded.

Some of the responses I got back were:

  * In our use-case family MPLS hash should apply. It is possible that
    MPLS hash computation algorithm may not recognized the packets in
    question.

  * Prior to 15.1R(2) the algorithm recognized up to two VLANs with TPID
    = 0x8100.

  * The TAC engineer told me that in MX platforms load balancing is very
    difficult to be accomplished equally across links.  Why? Because it
    works based on flows (same source MAC address, destination MAC
    address, source IP address, destination IP address, source port,
    destination port). This is very well known in MX, M and T series
    platforms (these are the ones I support).

  * Traffic imbalances may occur due to Insufficient variance in the
    packet flow. So this behavior seems to be working as designed.

So really, not much help, apart from the code changes in Junos 15.1 and
later.

I'd have had to get into a packet capture to drill deeper into the
issue, but this is a production box. The quagmire is that the affected
traffic is hyper-sensitive, but it is the one that would be useful with
a packet capture :-\.

Mark.



More information about the cisco-nsp mailing list