[c-nsp] WS-X6716-10G local switching and etherchanneling

[Tim Stevenson]

The 6708 is oversubscribed at the fabric, not at the port. But there 
are other limiting factors in the architecture. With 6708 you get 40G 
into the fabric, and up to 64G with local switching. But you won't 
get 80G out of this card.

HTH,
Tim

At 02:43 AM 7/3/2009, Sam Stickland contended:

>Thanks the reply Tim,
>
>Are the port's similarly oversubscribed on the 6708, or can line-rate be
>achieved between ports 1-4 & 5-6?
>
>Sam
>
>Tim Stevenson wrote:
> > Sam, please see inline below:
> >
> > At 04:38 AM 7/2/2009, Sam Stickland contended:
> >
> >> Hi,
> >>
> >> I've read:
> >> 
> <http://www.cisco.com/en/US/prod/collateral/switches/ps5718/ps708/prod_white_paper0900aecd80673385.html>http://www.cisco.com/en/US/prod/collateral/switches/ps5718/ps708/prod_white_paper0900aecd80673385.html
> >>
> >>
> >> If I'm understanding this correctly,
> >
> > I don't see any mention of 6716 in this white paper. 6716 does not
> > share the same architecture as any other 10G cards (eg 6708) mentioned
> > there. 6716 is actually more like a 6704 front ended by 4:1 muxes (at
> > a high level - in reality, different chips are being used, ie, metro &
> > r2d2 et al, not janus & rohini).
> >
> >> communication between each bank of
> >> 8 ports on a 6716-10G will be line-rate, but communication between the
> >> first and second groups of 8 ports will need to traverse the switch
> >> fabric?
> >
> > While it's correct that ports 1-8 & 9-16 are on separate fabric
> > channels, the key in the 6716 is that there is built-in *port-based*
> > 4:1 oversubscription.
> >
> > In other words, 4 physical 10G ports feed into a single 10G chip
> > (there are 4 such 10G chips on the card), ie, 4 ports share 10G of
> > bandwidth at the port level.
> >
> > So the maximum local switching performance you'd see in one half of
> > the card is 20G, the same as you'd get into the fabric.
> >
> >> On a similar note, if I create an etherchannel between two 6716-10G's
> >> will a module favour forwarding out of it's locally attached channel
> >> member?
> >
> > No, it's just a hash decision - luck of the draw. Eg, packet comes in
> > on t1/1 and channel member ports are t1/5 and t2/5. You've basically
> > got a 50/50 chance that you'd pass over the fabric.
> >
> > HTH,
> > Tim
> >
> >
> >
> >> Regards,
> >>
> >> Sam
> >> _______________________________________________
> >> cisco-nsp mailing list  cisco-nsp at puck.nether.net
> >> 
> <https://puck.nether.net/mailman/listinfo/cisco-nsp>https://puck.nether.net/mailman/listinfo/cisco-nsp
> >> archive at 
> <http://puck.nether.net/pipermail/cisco-nsp/>http://puck.nether.net/pipermail/cisco-nsp/
> >
> >
> >
> > Tim Stevenson, tstevens at cisco.com
> > Routing & Switching CCIE #5561
> > Technical Marketing Engineer, Cisco Nexus 7000
> > Cisco - <http://www.cisco.com>http://www.cisco.com
> > <http://www.cisco.com/>IP Phone: 408-526-6759
> > ********************************************************
> > The contents of this message may be *Cisco Confidential*
> > and are intended for the specified recipients only.
> >




Tim Stevenson, tstevens at cisco.com
Routing & Switching CCIE #5561
Technical Marketing Engineer, Cisco Nexus 7000
Cisco - http://www.cisco.com
IP Phone: 408-526-6759
********************************************************
The contents of this message may be *Cisco Confidential*
and are intended for the specified recipients only.