[c-nsp] Cisco 2800 performance at BRAS funcationality

[Oliver Boehmer (oboehmer)]

Robert Hass <> wrote on Sunday, February 19, 2006 10:50 PM:

> On Sun, 19 Feb 2006, Jim McBurnett wrote:
> 
>> Recently there have been a lot of performance questions..
>> 
>> So, I think this link is applicable:
>> 
>> http://www.cisco.com/warp/public/765/tools/quickreference/
> 
> According to below URL and routerperformance.pdf document my Cisco
> 3640 is able to forward 50-70 kpps. Currently is forwarding 10 kpps
> and CPU usage is nearly 100%. It means that specifications are
> different than real life applications. My applications using PPPoE
> VPDN termination, routing, OSPF and CAR rate-limit + uRPF on
> VirtualAccess interfaces. Probably these functions are Process
> Switching... 

No, those features should still be CEF-switched, but each of these
features (PPPoX encaps/decaps, CAR, uRPF) has an impact on forwarding
performance. "show proc cpu" should show most of the CPU usage be in
interrupt switching (the 2nd figure in 5 sec util being almost as high
as the first one).

> But what about ISR 2811 performance in 'Process Switching'
> (CAR+uRPF+PPPoE VPDN). Should I have performance like in Fast/CEF
> Switching (120kpps declared in specification) ?

The math in your case should be: The 2811 with its 120 kpps is about
twice as fast as the 3640 in the specific tests being run to arrive at
the stated numbers, so you will achieve about twice the performance for
your specific setup when you replace the 3640 with an 2811.. So you
might be better off with a faster platform than the 2811 to achieve a
more significant performance increase..

	oli