[c-nsp] full routing table / provider-class chassis

[Łukasz Bromirski]

On 2009-06-11 18:40, Kevin Loch wrote:

You've got something messed up Kevin:

> The DFC's only do next-hop (tcam) lookups and netflow.

DFCs are doing all and exactly the same work as PFC on
Supervisors locally on the LC that they are installed to.
They're the same in terms of hardware, just in a
different form - to fit the LC, not the Sup.

 > All packets are switched on the centralized PFC.

If the LC has a DFC, packet is switched by DFC toward
destination - if it's on the same card, it's switched locally
(until, of course, you seem to have 6748/6704/6708/6716 where
the card is divided into two).

If the LC doesn't have a DFC but CFC only, the traffic is
switched by PFC.

 > Each line card has two 20Gbit/s fabric channels
 > (2x 40Gbit/s full duplex) to the PFC.

Each 67xx series LC can have one or two 20Gbit/s channel
connections to switch fabric located at Supervisor. Switch
Fabric ASICs and PFCs are not the same thing.

65xx LCs have one or two 8Gbit/s connections to the switch fabric
and different DFCs models, but the switch fabric of Sup720/RSP720
can autonegotiate 8/20 Gbit/s upon insertion into chassis/boot.

> The PFC is rated at something like 30Mpps so if you are doing less
> than that and you don't need the extra netflow tcam you don't
> need any DFC's and can still theoretically do 640Gbit/s (320Gbit/s
> for those of us to have highly unbalanced traffic flows).

PFC is 30Mpps, DFCs for 67xx LCs can do 48Mpps.

-- 
"Everything will be okay in the end. |                  Łukasz Bromirski
  If it's not okay, it's not the end. |       http://lukasz.bromirski.net