[c-nsp] Per-packet load-sharing on hardware based platforms.

[Dan Holme]

>> I didn't realise until today that you can't do CEF per-packet
>> load-sharing on hardware-based platforms, ie. 6500/7600, this is
>> purely because I haven't tried to implement it until now. The traffic
>> I'm trying to load-balance is a single flow, encapsulated inside a
>> port-based pseudowire, therefore I was going to use per-packet
>> load-sharing on two MPLS TE tunnels destined for the endpoint.
>>
>> What workarounds or alternatives can people suggest?
>
> Hi Dan,
>
> With MPLS-TE you can configure the tunnel to weight CEF hash buckets.  The command you are looking for is 'tunnel mpls
> traffic-eng load-share #' configured under the tunnel interface.  I believe you can use 'show ip cef x.x.x.x internal' to see the
> hash-bucket allocation for a given prefix.  It seems to me it didn't balance properly with 12.2(18)SXF; we always saw 50/50
> load regardless of weighting.  However, it did provide the desired results running 12.2(33)SRC.  I stopped using it a year or
> two back, so you'll probably have to check the IOS train features relevant to your kit.  I'm only using MPLS-TE for FRR at
> this point.
>

Ah forgot about the TE load-share... I fiddled around with it before
and never got it to work, but truth be told I have only tried with SXF
so far. I'll let you know my results from testing with SXH/SXI.

-- 
Regards,
Dan Holme