[c-nsp] 6500 Sup720 high CPU load - RP LES Fragmentation unsupported

[Bernhard Schmidt]

Hi,

we have a Cisco Cat6500 with Sup720a running 12.2(33)SRA. This box is
experiencing high CPU load in the IP Input process:

csr1-2wr#sh proc cpu sorted 
CPU utilization for five seconds: 54%/27%; one minute: 63%; five
minutes: 66%
 PID Runtime(ms)   Invoked      uSecs   5Sec   1Min   5Min TTY Process 
 131   100248440  50432579       1987 23.19% 27.28% 28.14%   0 IP Input         
   7     3076072   6211647        495  0.71%  0.63%  0.62%   0 ARP Input

"sh ip cef switching statistics" shows a high number of packets punted
to RP because of "RP LES Fragmentation unsupported". "sh int stats"
shows 99% of the packets being in the Processor Switching path come from
one routed TenGigabitEthernet-Interface. Most traffic coming in there
should be routed to a SVI. On the other side of the 10GE link is another
6500 Sup720a (running 12.2(18)SXD7), which has only two
Layer3-Interfaces (the 10GE in routed mode and one SVI).

I could not find anything at cisco.com about "Fragmentation
unsupported". Someone in IRC suggested it might be due to the first 6500
fragmenting IP packets coming in without the DF bit set. This could be a
problem, as the 10GE routed interfaces are set to 9216 bytes and the L3
SVI are set to 1500 bytes.

------ Vlan20 SVI - Te2/4 routed ------- Te1/1 routed - Vlan71 SVI
         (1500)       (9216)               (9216)         (1500)
	         ^^^ CPU-Load

Te2/4 shows quite a number of giant packets coming in. Problem is, there
should be none, as the second box only has one IP interface which is MTU
1500. Also it does not show any giant packets on any physical interface.
The Traffic on Te2/4 suggests that the packets are not jumbos (25Mbps in
3kpps), but that could be a bad mix as well.

Can someone verify that "RP LES Fragmentation unsupported" are indeed IP
packets being fragmented because they are sent on a link with a smaller
MTU?

Thanks,
Bernhard