[c-nsp] Getting ready to pull the trigger: RSP720/SUP720

[Peter Rathlev]

On Mon, 2009-04-06 at 07:51 -0700, Rick Ernst wrote: 
> The problem I am running into is spec'ing the aggregation layer.
> Almost all of our traffic is ethernet now, and all the interfaces need
> bi-drectional rate-limiting/traffic-shaping/policing.  We have a
> variable bandwidth model and need to cap traffic at 1Mbs granularity.
> 1,5, and 10Mbs connections are common, and 20,50,100Mbs connections
> exist with a 200Mbs pipe in process.

ES20 line cards for the 7600 might fit your purpose:

http://www.cisco.com/en/US/prod/collateral/routers/ps368/product_data_sheet0900aecd8057f3ad.html
http://tinyurl.com/3jc8mx

The hardware forwarding switches can't really shape very well on LAN
card interfaces. You can use token bucket policing, and the "SRR"
enabled interfaces can do some crude "timeslot" shaping, but real
buffered shaping isn't possible AFAIK.

On Mon, 2009-04-06 at 11:12 -0400, Jon Lewis wrote:
> On Mon, 6 Apr 2009, Rick Ernst wrote:
> > I'm planning on collapsing the border/core into a pair of
> > 7600/Sup720-3BXLs, and it looks like they will be almost idle with
> > this amount of load.
> 
> That really depends on the features you enable.  Try doing full
> netflow on a sup720 doing a few hundred mbit's of traffic, and they're
> suddenly not so mighty.

Sorry if I repeat myself, but I don't understand this problem. We export
netflow from Sup720-3Bs often carrying >1 Gbit/s and the processor
hardly seems to notice. And it's with an if-full flowmask. We're still
on SXF but I sincerely hope SXH/SXI behave the same way for us.

Am I missing something here?

Regards,
Peter