[c-nsp] GRE on Cat-4948 switch

[Harald Firing Karlsen]

Justin Shore wrote:
> Ian MacKinnon wrote:
>> Hi All,
>>
>> Does anybody have any idea the impact of running multiple GRE tunnels 
>> on a 4948 switch?
>>
>> I can see that it will be processed by software rather than hardware, 
>> but just how much of a problem is this likely to be?
>>
>> I am only talking about a max of 100M of GRE traffic, amongst a 
>> couple of Gig of total traffic.
Thats impossible to say without knowing what kind of traffic. If it's a 
lot of small packets you will get a high PPS and thus a high overhead 
which causes high CPU load. The 4948 might be able to do 100Mbit of 
GRE-traffic under ideal conditions (only 1500B+ packets, etc), but 
without knowing/controlling the traffic 100% it is very risky.
>
> I think one of the actual cisco.com guys would have to speak up to 
> answer this question properly with figures on CPU speed, processing 
> times, etc.  I would say with an off-the-cuff figure is that process 
> switching 100M would swamp most CPUs that Cisco has on the market.  
> I'm fairly confident that our Sup720-3BXLs wouldn't take kindly to 
> it.  The experts would probably have a better answer for you though.
Well, the SUP720-3BXL does GRE in hardware, so pushing 100Mbit+ on a 
SUP720-3BXL shouldn't be a problem.

-- 
Harald Firing Karlsen