[c-nsp] Cisco 6509 and Bus speeds

[Tim Stevenson]

See below:

At 08:18 AM 1/13/2005, cisco-nsp-request at puck.nether.net averred:
>Message: 4
>Date: Thu, 13 Jan 2005 13:45:03 +0100 (CET)
>From: Marcel Lemmen <marcel at support.net>
>Subject: [c-nsp] Cisco 6509 and Bus speeds
>To: cisco-nsp at puck.nether.net
>Message-ID: <Pine.LNX.4.55.0501131323160.16860 at td.office.support.nl>
>Content-Type: TEXT/PLAIN; charset=US-ASCII
>
>Hi all,
>
>My company is looking for a replacement of our Cisco 4006 and we're
>looking at the 6509 with Sup720. I do however have some questions
>regarding the Bus and backplane speeds which are a bit unclear to me.
>Perhaps somebody could help me with this one.
>
>We are looking for a configuration with a Sup720, 2x 4 port 10GE
>line-cards, a 48port 10/100/1000 and a 24port 1GigE SFP.
>
>I have some nasty experiences with promised performance and actual
>performance, like a 7500 with GEIP+ witch only handles 400Mbit/s, a 4006
>switch with 3x24Gbit/s SwitchEngines and gives packetloss at 1.8Gbp/s etc.
>
>The c6509 with Sup720 can handle 720Gbps of switching traffic in total,
>but what happens if I transfer 40Gbps from one slot to the other slot with
>4x10GE (since we have 8 10GE ports)?

This card is not fully non-blocking, but it is nearly (like >98%) line rate 
on all 4 ports with large packets. You can't ever get full 40G wire 
ethernet across the fabric, as we add 32 bytes of overhead to every packet. 
Also, the fwding engine is the bottleneck at 64B packets, ie, either the 
sup (30Mpps) or a DFC3 (48Mpps) has less capacity than 4x10G ports (60Mpps).

>  As far as I know the c6509 has a
>8Gbps backplane

It is a 16G (32G marketing) bus. If you don't have DFCs, it is used to xfer 
packet headers to the sup for lookups, with DFCs, it is not used. In either 
case, the fabric channels, not the 16G bus, are used to xfer the actual 
packets/payload.

>and therfor you shouldn't be able to transfer more then
>8Gbps from one slot to another, or am I wrong in this one?

Wrong, with the cards you mention you have 2x20G (40G marketing) channels 
to each card (40G per slot, 80G marketing), except the 6724 (1x20G 
channel). These all interconnect thru the fabric on the sup.

>Using dCEF it
>would be possible to transfer 40Gbps of traffice within one slot but not
>between slots or not?

With DFC3, you can get (nearly) 40G from slot to slot. Nearly, again due to 
that 32B of overhead per packet.

Tim

>Could someone explain this to me because I want to know a bit more about
>the architecture so I won't make the same mistake as I did earlier :)
>
>Thank you in advance!
>
>With kind regards,
>
>Marcel Lemmen



Tim Stevenson, tstevens at cisco.com
Routing & Switching CCIE #5561
Technical Marketing Engineer, Catalyst 6500
Cisco Systems, 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.