I have question about the Nexus 5596 with the Layer 3 daughter card. Looking at the specs of the 5596UP:
Cisco Nexus 5596UP and 5596T: Layer 2 hardware forwarding at 1920 Gbps or 1428 mpps; Layer 3 performance of up to 160 Gbps or 240 mpps.
When looking at the hardware it was my understanding that if I have L3 card I would remove the L2 card from the Nexus the Layer 2 I/O module (N55-DL2) and replace it with the L3 card.
Here is my question if I am running in L3 mode do I loose the forwarding at L2, the 1920Gbps and can only get up to 160 Gbps?
Or are these separate, L2 forwarding rate is still 1920 Gbps and any L3 traffic rate is 160 Gbps?
With a combined throughput 2080 Gbps?
Thanks for you help.
The simple answer is that 1920Gbps is the maximum bandwidth supported by the switch fabric. When you install the Layer-3 module it will use some of that 1920Gbps rather than add to it.
The slightly longer answer...
The 1920Gbps that Cisco quote is effectively the crossbar fabric bandwidth. If you look at the attached image from a Cisco Live presentation (BRKARC-3542) the Fabric is shown in the centre as the Unified Crossbar Fabric (named Sunnyvale) and this has a number of port ASICs shown as UPC (Unified Port Controller) that connect to it.
On the Nexus 5596 you have up to 96-ports running at 10Gbps which can all get full-duplex non-oversubscribed access to the switch fabric via one of twelve port ASICs. So you have 96 ports * 10Gbps * 2 (full duplex) which comes out at the 1920Gbps that's quoted.
As you say, when you install the Layer-3 module you use one of the expansion module slots. If you look at the top left of the architecture diagram you can see a box around two of the port ASICs which is marked as Expansion Module. The Layer-3 expansion module connects to the fabric via two ports ASICs, and so uses a total of sixteen port ASIC connections into the switch fabric. As each of these are 10Gbps we arrive at the 160Gbps Layer-3 performance that is quoted. Note that even though these connections are full duplex a single routed packet uses the bandwidth in both directions as the packet is looped within the Layer-3 module. The other thing to note is that the 160Gbps quoted is actually a 16 * 10Gbps port-channel, with traffic balanced across the port-channel on a session basis. So as with any port-channel, the maximum bandwidth any single traffic stream can use is limited to the capacity of one of the physical links, which in this case is 10Gbps.
The Cisco Live presentation is obviously public so if you want to understand packet forwarding etc., for the platform it's probably worth doing a search for BRKARC-3542, either on Google or on the Cisco Live 365 site if you've registered.