Core issue
These are the two causes of high CPU utilization:
- Software forwarding. The software forwarding causes the CPU utilization to go high because the SDM prefer command is issued as the desktop. The desktop does not prevent other features from using the memory allocated to unicast routing. So the memory is used by other features. This makes the software process the forwarding activities due to lack of memory to routing.
- GMPSN and HRPC h2lmm request processes. HRPC is a communication tool between stacked switches about Layer 2 (L2) features, and IGMPSN is Internet Group Management Protocol (IGMP) snooping. If IGMP packets are bouncing between stacks, those two processes can cause high CPU. You can check this by issuing the debug ip igmp snooping command.
A storm of IGMP leave messages is sent between stacks of Cisco Catalyst 3750 switches. This problem can occur when there are two or more stacks of Cisco Catalyst 3750 switches connected to each other through non-routed (L2) ports, and a client sends an IGMP leave message.
Resolution
To resolve the issue, perform one of these procedures:
- Issue the show controller CPU command to determine if the CPU utilization is due to the software forwarding. Look at the swf forwarding field. If the CPU utilization is due to the software forwarding, issue the sdm prefer routing command in all the Catalyst 3750 switches in the stack, and reboot the stack.
- The workaround is to disable IGMP snooping globally or on a VLAN, where the IGMP leave flooding is seen. Issue the no ip igmp snooping command.
This issue is documented in Cisco bug ID CSCeg57882. This issue is fixed in code versions 12.2(25)SEB, 12.2(25)SEA, 12.2(20)SE4 and later.
