[c-nsp] 6500 xconnect port-channel?

[Benjamin Lovell]

>From a load sharing point of view this is likely to work out better as then the CEs have all the load-sharing options available to them. LACP, STP, UDLP, et al pass transparently over port mode xconnect. 

-Ben

On Jul 14, 2010, at 12:13 PM, Peter Rathlev wrote:

> On Wed, 2010-07-14 at 09:49 -0400, Benjamin Lovell wrote:
>> I would test this for load-sharing with your traffic profile. It's
>> been a while since I looked at this but I think the port channel
>> hashing criteria is limited when xconnect terminates to a
>> port-channel. 
>> 
>> If I remember correctly we cannot do port channel hash on IP header
>> when coming off an EoMPLS tunnel. So if all/alot of traffic has same
>> MAC then you will get sub-optimal load-sharing. 
> 
> I see. Maybe a better idea is to create to independent port-mode ports
> and then let the CE-devices run LACP through these? Each side could the
> load-share as they like, and that would stay closer to the "pseudo-wire"
> concept I guess.
> 
> This is for inter-DC L2 connections, and we deliver two distinct EoMPLS
> connections (traffic engineered to use completely different paths and
> networking devices) that the CE-devices run Rapid-PVST on top of. The
> idea was to give them 2 x 2 Gb/s connections instead of 2 x 1 Gb/s, seen
> from the CE as two port-channels. (The two members of each port-channel
> would use the same TE LSP, since it's just for more bandwidth.)
> 
> -- 
> Peter
>