SNMPV3 - Zabbix Server CPU nice high

[jrprado]

Staff, checked a problem adding three SNMPv3 devices in Zabbix, CPU consumption and increased detail is very nice which is high. If I add more devices to V3 CPU is 0% idle. Here some screens with evidence of increased consumption.

TOP


Graph CPU


Latest Data


Status of Zabbix


NOTE: All devices are collecting the data correctly and I just changed from SNMPv2 to SNMPv3 making the appropriate changes.

What can be done?

[Blakkheim.GW]

Hmm, this is weird. We have more than 60 hosts (which represents more than 2000 items) checked by SNMPv3 and we have no performance problem.

What v3 method do you use ? authNoPriv ? authPriv ? If authPriv is used, maybe it causes an overload.. I only use authNoPriv, so I can't say.

How many pollers did you start in your zabbix_server.conf ?

Cheers.

[jrprado]

Hmm, this is weird. We have more than 60 hosts (which represents more than 2000 items) checked by SNMPv3 and we have no performance problem.

What v3 method do you use ? authNoPriv ? authPriv ? If authPriv is used, maybe it causes an overload.. I only use authNoPriv, so I can't say.

How many pollers did you start in your zabbix_server.conf ?

Cheers.

Hi Blakkheim.GW, thanks for the reply.

I'm using authPriv method. Is that what is causing the high CPU consumption? Can you tell me the difference in methods? I will search and then modify.

I do not think that is the setting Pollers, but would occur delay in the queue. Here's my setup zabbix_server.conf:

StartPollers=30
StartPollersUnreachable=10
StartTrappers=30
StartPingers=10
HousekeepingFrequency=24
CacheSize=16M
StartDBSyncers=4
HistoryCacheSize=8M
TrendCacheSize=12M
HistoryTextCacheSize=16M

[jrprado]

I performed some tests changing the security level. As you will see in the chart below, there was a reduction of 60% to 50% with AuthNoPriv and a reduction of 50% to 30% with NoAuthNoPriv. The difference for the AuthPriv NoAuthNoPriv was 30%.



I will check with the team responsible for networking if you can change the settings of the switches.

In Zabbix is anything I can do if I have to monitor for AuthNoPriv? I'm currently with 3 switches with SNMPv3 and still got 40 switches to migrate to SNMPv3. As it stands the server will be unresponsive due to high CPU consumption.

My server configuration:
2x CPU
model name: AMD Opteron(tm) Processor 852
cpu MHz: 2600.000
Mem: 4059036 kB

[Blakkheim.GW]

Huumm.. Monitoring such few items with authNoPriv method shouldn't lead to such CPU consumption.. My own server is quite powerful so I can't say if this is a normal behaviour but it seems excessive to me.

Do you use static OIDs ? Dynamic ? Do you use numeric OIDs ? Or the textual ones ?

Keep in mind that if you use NoAuthNoPriv, you loose all the advantages of v3.. So maybe you should consider keep using v2..

[jrprado]

Huumm.. Monitoring such few items with authNoPriv method shouldn't lead to such CPU consumption.. My own server is quite powerful so I can't say if this is a normal behaviour but it seems excessive to me.

Do you use static OIDs ? Dynamic ? Do you use numeric OIDs ? Or the textual ones ?

Keep in mind that if you use NoAuthNoPriv, you loose all the advantages of v3.. So maybe you should consider keep using v2..

The advantage in SNMPv3 is not only greater access control? Or is more?

Below is the configuration of items:

[jrprado]

No answer...

My settings are proper. Is there anything I can do?

Is that the problem is the hardware of my server that is not enough to monitor SNMPv3?

I thank you for the help.

[Alexei]

SNMPv3 requires significantly more CPU power due to additional cryptography related computations Zabbix has to deal with. Not sure how powerful your CPUs are, yet I wouldn't expect such a massive performance hit after switching to SNMPv3 in your case.

[jrprado]

SNMPv3 requires significantly more CPU power due to additional cryptography related computations Zabbix has to deal with. Not sure how powerful your CPUs are, yet I wouldn't expect such a massive performance hit after switching to SNMPv3 in your case.

Hi Alexei, thanks for the reply.

The difference in CPU consumption due to the change of snmp v3 was great. The version of snmp package installed would have any influence? Following packages are installed:

net-snmp-5.3.2.2-5.el5
net-snmp-perl-5.3.2.2-5.el5
net-snmp-devel-5.3.2.2-5.el5
net-snmp-libs-5.3.2.2-5.el5
net-snmp-utils-5.3.2.2-5.el5

OS Description: Red Hat Enterprise Linux Server release 5.3 (Tikanga) 64 bits

At the moment I do not have other hardware to test to see if the same thing happens.

[Alexei]

The difference in CPU consumption due to the change of snmp v3 was great. The version of snmp package installed would have any influence?

I do not think so. It looks like you're collecting lots of data (low update interval?), otherwise I cannot explain the high CPU load.

[Blakkheim.GW]

I agree with Alexei. Trying to grow your intervals between checks can help. I can't see any other explanation.

Cheers.