Problem Details: I have two switches: Nexus 3064 (ver 5.0(3)U1(2)) and Cisco 6509. The 6509: 1 x WS-x6708-10GE 8 port Ten Gig module version 3.5, firmware 12.2(18r)S1. 2 x sup 720's, with PFC3A. 2 x WS-x6348-rj-45 4 x WS-x6748-ge-tx The IOS is ipservicesk9-mz 12.2(33)SXH8b. Both switches have been running fine for quite a while(not connected to each other). I then ran a fiber connection between port Nexus:1/48 and 6509:Te9/1 The Nexus is configured for jumbo frames: policy-map type network-qos jumbo class type network-qos class-default mtu 9000 system qos service-policy type network-qos jumbo interface Vlan323 no shutdown mtu 9000 ip address 172.19.4.253/24 ip pim sparse-mode (all 48 ports are configured the same) interface Ethernet 1/48 switchport access vlan 323 mtu 9000 The above works on the Nexus switch, I have no problems with jumbo frames inbetween Nexus users. On the 6509: system jumbomtu 9000 interface Vlan 323 ip address 172.19.4.254 255.255.255.0 ip pim sparse-mode mtu 9000 interface TenGigabit9/1 switchport switchport mode access switchport access vlan 323 mtu 9000 When I ping (of any packet size) from the Nexus to the 6509 @172.19.4.254, the 6509's CPU goes to 100%. On occasion, we will get 1 out of 20 packets back in the reply. I reduced the MTU size on the three 6509 parameters until the CPU stopped going 100%.
The magic number is 4175 bytes. 4176 and higher == 100% CPU.
I am willing to put the fiber link to 1500, but how does one change that for just one port on the Nexus? I tried and it refuses to set the mtu. I also tried to setup a new service-policy but that didn't seem to work either. The Nexus users are all jumbo frame users. The users on the 6509 are all 1500 byte frame users, except for one user on a 1 gig port (ws-6748-ge-tx) line card. It is this user whom would like jumbos. I'm more worried about why the 6509's CPU went to 100%! Any idea? thanks! Mark
p.s. I attempted to put in a TAC request, but apparently my devices are not covered under SmartNet
Are you running sup 720 or 32? What else you have conf on the 6509?If you have a mile long reflexive acls, extended QoS, or powerful servers feeding in the switch, that's alone can shotup cpu over 75-80%.. Did you check on the fans? Do a sh env and make sure fans are cooling off the modules. Also check cooling in MDF is enough.I don't think the 6748 moduless are "made" to support jumbo frames(not withstahding what Cisco say otherwise).
Thank you for the reply.
We are running 720's with no ACLs. Normally the 6509 runs at 1% CPU utilization. The CPU only splikes to 100% when we enable jumbo frames past 4175 bytes. We're not even passing traffic yet, just setting the jumbo mtu causes the spike.
Interesting information regarding the 6748's possibly not supporting jumbo frames... I'll have to do some experiments.