[f-nsp] NetIron MLX Experience..

Richard A Steenbergen ras at e-gerbil.net
Tue Aug 8 17:50:56 EDT 2006


On Tue, Aug 08, 2006 at 02:09:16PM -0700, Tom Samplonius wrote:
> 
>   Strictly speaking, Cisco CEF is just a way to store a forwarding table 
>   for fast lookups.  The storage could be in some kind of content addressable 
> memory, and looked up by an ASIC, or it could be in RAM and looked up by 
> the the router's CPU.

CEF originally implied a software data structure (an mtrie) which was 
dedicated to doing forwarding lookups only (a FIB), and which was 
pre-populated to deliver deterministic lookup speeds whether the packet 
was the first in a flow (or first to a destination) or not. Some later 
techniques in hardware looked pretty much like this, others did not, but 
the marketing term stuck. :)

>   So Foundry made the jump to putting the full routing table in CAM with 
>   the MLX?  Juniper has always operated this way, so why have some 
> respondants talked about replacing Juniper's with MLXs?  Other than that 
> some of the old processing engines on the Juniper's lacked enough memory 
> for some features, there should be little difference at the L2/L3 
> forwarding level these days?

Sure, but prepopulating the CAM with a full table is "building a router 
101", and has almost nothing to do with the success or failure of 
replacing a Juniper. Like I said before, doing a bazillion lookups/sec or 
moving a bazillion bits/sec is trivial, in L2 or L3. It is a problem that 
has been solved, the hw is a commodity. The difference between a crappy 
box and a good box is software, software, software.

-- 
Richard A Steenbergen <ras at e-gerbil.net>       http://www.e-gerbil.net/ras
GPG Key ID: 0xF8B12CBC (7535 7F59 8204 ED1F CC1C 53AF 4C41 5ECA F8B1 2CBC)



More information about the foundry-nsp mailing list