[nsp] Re: Catalyst 6500 Hybrid

[Mikael Abrahamsson]

On Sun, 2 May 2004 sthaug at nethelp.no wrote:

> I don't see how the Sup720 is more of a "real" router than Sup2, with
> the possible exception of the TCAM size (enabling uRPF with current
> Internet routing tables). The Sup2/MSFC2/PFC2 still do longest prefix
> match like any other router.

Yeah, I'm not much into the 6500 specifics.

> Interesting that you should mention buffer sizes. On the end2end-interest
> there is now an argument going, part of which is that you probably do
> *not* want as deep buffers today as you previously did. However, 80ms
> would still be nice to have!

Most people run WRED etc anyway, you never want 600ms latency even on a 
full link. Running single TCP sessions require this, but in effect if 
you're running thousands of sessions over a full 10GE, running WRED or 
having small buffers to begin with, makes marginal difference, you get the 
same effect anyway-
 
> Agreed that 12000 and Junipers are expensive - the basic impression is
> that any kind of equipment that can handle hardware based forwarding 
> for non-Ethernet media is outrageously expensive. I don't see this
> changing any time soon :-(

The worst part is that the per megabit price on ports hasnt gone down on                            
these platforms for 5 years (or only marginally so).                                                
                                                                                                    
A STM64 port is almost exactly 4 times the price of STM16 and has been                              
like that for a long while. 

If you need distributed forwarding with subsecond route table switches, 
deep buffers and all that, go GSR or Juniper, I hope you're able to charge 
your customers for the premium service you're offering them. For internet, 
it being "best effort" and all, I dont see customers wanting to pay extra 
for this.

-- 
Mikael Abrahamsson    email: swmike at swm.pp.se