[c-nsp] qos (?) capacity question

[John Gill]

Hi Tom,
Mikael's guess is very likely the way it would be implemented if they 
are using a 6500.  The nickname for this type of microflow policing is 
UBRL - user-based rate-limiting.

It is limited based on the PFC/DFC's in use and the flowmask they are 
using.  For example, if you identify one "user" as a source IP address, 
then 1 flowmask is needed to police that user.  If you identify a user 
as source destination IP address, now all of those flows create new 
entries and the table fills more quickly.  UBRL is usually just source 
IP so it can grow fairly nicely, depending on what other features are 
using Netflow and if they can move to a source flowmask.  For instance, 
if the 6500 is responsible for NAT, a source only flowmask would not be 
possible.  NAT also uses the netflow table, and a handful of other 
features do too.

 From Table 2 here: 
http://www.cisco.com/en/US/prod/collateral/switches/ps5718/ps708/prod_white_paper0900aecd803e5017.html

Sup720 effective capacity 64k flows
Sup720-3B or 3C effective capacity 115k flows
Sup720-3BXL or 3CXL effective capacity 230k flows

This would apply *per* DFC or PFC.

I hope that helps.

Regards,
John Gill
cisco

On 5/31/11 9:09 AM, Mikael Abrahamsson wrote:
> On Tue, 31 May 2011, Tom wrote:
>
>> we are doing lab training on univ. with some cisco 6509, force10 e300
>> and juniper m.
>
>> Is this type of policing implemented by qos?
>
> Cisco 6500/7600 has something called "microflow policing" which might do
> some of what you're describing.