[c-nsp] x6148 vs. x6548

[Michael Ulitskiy]

On Friday 12 June 2009 11:34:14 am Pete Templin wrote:
> Geoffrey Pendery wrote:
> 
> > I have a question of my own, since this subject has come up a time or
> >  two - regarding the 6148's, the statement is made a couple times
> > that Etherchannel will get you port redundancy but no extra
> > bandwidth, since the ASIC is only a gig.  But if I distribute my
> > channel across two slots, say Gig 1/1 and Gig 2/1, does that get me
> > around the gig limit?  Or even Gig 1/1 and Gig 1/48, since it's
> > separate ASICs? Logic tells me yes, but I've heard the "1 gig limit"
> > mentioned as if it's a hard platform limitation, not just a result of
> > a particular bottleneck.  My instinctive behavior with channels is to
> > span them across blades anyway, to guard against blade failure....
> 
> My understanding (since my google-fu won't find a quickie answer at the 
> moment) is that 6148s copy any EtherChannel frames to every ASIC on the 
> card, so you can get to 2G by spreading over two cards, but you're still 
> limited to 1G no matter no many controllers you cover within a 6148.  :(
> 
> We've updated the banners on all relevant 6148-loaded chassis to remind 
> folks to never build EtherChannels on those cards.  Oh well...

My understanding was that every EtherChannel frame is delivered (by Sup) to every
ASIC involved (has a port) in EtherChannel regardless of which card it is on.
So you can't get more than 1G even if you distribute your EtherChannel over several cards.
Am I wrong?

Michael