WS-C3750X stack loosing management access once SNMP is enabled

[HeliosX]

Hi,

I have a stack of 5 WS-C3750X switches which work fine but when I enable SNMP and then set up poling on Observium I immediately loose remote access to it. The only way to fix it is to disable poling and do a cold reboot.

Does anyone have any solution for it so I can have SNMP working on it ?

This is the setup:
1 18 WS-C3750X-12S   15.2(1)E1    C3750E-UNIVERSALK9-M
* 2 30 WS-C3750X-24   15.2(1)E1    C3750E-UNIVERSALK9-M
3 18 WS-C3750X-12S   15.2(1)E1    C3750E-UNIVERSALK9-M
4 18 WS-C3750X-12S   15.2(1)E1    C3750E-UNIVERSALK9-M
5 18 WS-C3750X-12S   15.2(1)E1    C3750E-UNIVERSALK9-M

[HeliosX]

I do not really need all the fency features of 15.2 version so I will try 15.0(2)SE12 as you suggested Joseph. Thx.

[HeliosX]

Andre,

Im planning on upgrading software in that stack.

In the meantime I found a temporary solution. 

I have configured snmp-server on the switch one more time and then added it to Observium while monitoring CPU usage.

When Observium started polling information the CPU once again went to 99% and stayed there. I then removed the snmp config from the switch while leaving added device in Observium.

After a few minutes I have added snmp-server config to the switch once more while monitoring CPU utilisation.

This time the SNMP Engine did not take over all the CPU resources. Few times it went to 30-40 % but then it stabilised.

I believe it must be one of the Discovery Modules in Observium that cisco doesn't like because when a device is first added to Observium it uses all discovery modules available and then once a device type is discovered it only uses some of them...

[andrew.butterworth · Accepted solution]

There is a bug listed regarding high CPU with snmpwalk (CSCum72168) so I wonder if the SNMP server is walking through all MIBs to discover the device?

Does the 3750X support control-plane policing?  Might be worth looking into that?

[HeliosX]

Thx Andrew for that bug report. I have now disabled cisco-cbqos discovery module in observium and hopefully it will be ok now.  Will let you know.

[HeliosX]

Disabling cisco-cbqos discovery module in Observium seems to have solved the problem. The CPU is not going to max anymore on the discovery.

[HeliosX]

Andrew,

Im planning on upgrading software in that stack.

In the meantime I found a temporary solution. 

I have configured snmp-server on the switch one more time and then added it to Observium while monitoring CPU usage.

When Observium started polling information the CPU once again went to 99% and stayed there. I then removed the snmp config from the switch while leaving added device in Observium.

After a few minutes I have added snmp-server config to the switch once more while monitoring CPU utilisation.

This time the SNMP Engine did not take over all the CPU resources. Few times it went to 30-40 % but then it stabilised.

I believe it must be one of the Discovery Modules in Observium that cisco doesn't like because when a device is first added to Observium it uses all discovery modules available and then once a device type is discovered it only uses some of them...