[c-nsp] Shapping in/out on Ethernet subinterfaces (7206 NPE-G1) and CPU usage

[Gauthier DOUCHET]

Hello all,

On a 7206 NPE-G1 which is an access router for customer, we want to do some
shapping.

At this moment, we are using the configuration just described below.
In this case, customers are directly connected to the gi0/2 router (through
a switch).

Is it the best way to do about the subint configuration?

Moreover, we added 20 subinterfaces in this situation recently.
At total on this router, we have 100 subinterfaces (half are in a vrf for
private subnets and natting to Internet, other are just routing public
subnets).
We saw the CPU usage growing in a significant way (75% CPU).
Is this due to a bad configuration of the subints or just the fact the
NPE-G1 is a software router too weak for this kind of usage?
Would be a NPE-G2 sufficient on short term before moving to a real router
as a ASR 1k?

Thank you for help,
Regards
Gauthier

class-map match-any CLASS_ALL_TRAFFIC-10M-UPLOAD
 match any
class-map match-any CLASS_ALL_TRAFFIC-10M-DOWNLOAD
 match any

policy-map SHAPING_ALL-10M-DOWNLOAD
 class CLASS_ALL_TRAFFIC-10M-DOWNLOAD
  shape average 11080000

policy-map POLICING_ALL-10M-UPLOAD
 class CLASS_ALL_TRAFFIC-10M-UPLOAD
  police 11080000


interface GigabitEthernet0/2.21
 description Cust1 [10Mbit] {R0012}
 encapsulation dot1Q 21
 ip address X.X.X.X 255.255.255.252
 no ip redirects
 no ip unreachables
 no ip proxy-arp
 ip virtual-reassembly in
 ip verify unicast reverse-path
 no ipv6 redirects
 no ipv6 unreachables
 ipv6 verify unicast reverse-path
 ipv6 virtual-reassembly in
 no cdp enable
 service-policy input POLICING_ALL-10M-UPLOAD
 service-policy output SHAPING_ALL-10M-DOWNLOAD

-- 
Gauthier DOUCHET