Nexus 7000 CPU Utilisation

[j-mccarthy]

Hi

I have a pair of Nexus 7000s acting as distribution switches for 52 access layer stacks of 3750x's.

Southbound there are 52 Layer 2 VPCs into each N7K. The N7Ks are a HSRP pair providing L3 gateway functionality for all devices connected to the access layer stacks.

Northbound is Layer 3 Routed using EIGRP.

I was running 5.1.2 on both, I have now upgraded one of them to 5.2.1 but the behaviour is the same.

N7K#show process cpu history

    767097676567655666567677666666065706566767666665676764666665675656675767
    293035610892288400697302942919049406951470983107411038553524005029828304
100    *                          *   *
90    **                         *   *
80    ***              *         *   *
70 ********  **       ** ***  * **  *** * *****     * *  ** *   *   *** * *
60 ***************************************************** ***** **** *******
50 ************************************************************************
40 ************************************************************************
30 ************************************************************************
20 ########################################################################
10 ########################################################################
    0....5....1....1....2....2....3....3....4....4....5....5....6....6....7.
              0    5    0    5    0    5    0    5    0    5    0    5    0

                   CPU% per hour (last 72 hours)
                  * = maximum CPU%   # = average CPU%

I'm concerned about the CPU peaking at 100% and averaging at 20%, seems a bit high to me. Is this normal, or should I open a TAC case?

N7K# show system resources
Load average:   1 minute: 0.05   5 minutes: 0.25   15 minutes: 0.29
Processes   :   344 total, 2 running
CPU states  :   12.5% user,   15.0% kernel,   72.5% idle
Memory usage:   4109560K total,   1586740K used,   2522820K free
Current memory status: OK

[dominic.caron]

Hi,

This high cpu will not give you any problem on your trafic, the 7000 platform as a robust preemptive scheduler and the control and dataplane are seperated.

Depending on what you do, the cpu might be a bit high. High cpu is cause be misbehaving control protocol or excessive trafic to the cpu.

You can start by looking at what process is taking a lot of ressource. Filtering on that process, try capturing cpu bound trafic using the internal Wireshark.

Could be arp, glean, dhcp,fragment....need a sh process cpu sort

[Jerry Ye]

The periodic CPU spike in the N7K is normal (usually caused by a process called platform). It is because the N7K will perform periodic internal function and this function generate a CPU spike.

HTH,

jerry

[j-mccarthy]

I'm running an SNMP access list and also have the default COPP strict priority, pretty surprised anything external could push the CPU to 100%  with the strict COPP policy applied..

[HUBERT RESCH]

We have 2 N7K in a vPC Bundle , the one which is HSRP standby has significate highe cpu , about 40% higher caused by process platform.

Now we are investigating what process platform could make it to consume such hoch cpu load

[sbhadrav@cisco.com]

Virtual Port-channel (VPC) is a feature for Nexus-connected L2 networks where a pair of Nexus devices share the data-plane of LACP port-channels while maintaining separated management and control planes. This yields a loop-free L2 trunk connection between access/distro and core layers granting additional non-blocking uplink bandwidth.
Downstream L2 devices communicating with gateway HSRP SVI’s over VPC has never been a problem on Nexus. Static routing and L3 routed P2P have also worked since routing was introduced on Nexus.
Dynamic routing with EIGRP/OSPF/BGP, however, between the VPC-connected Nexus platforms and another device on a L2 VLAN segment requires very careful planning. If not handled properly, a misconfigured Nexus dynamic routing peering will cause network instability.