Inconsistent results?

[jrdbeau]

Hi,

I'm getting some inconsistencies when I add identical systems to my device list. I have 2 identical Windows Server 2003 systems sitting right next to each other and I get different results within Cacti. Both systems are running SNMP Informant Std and I am using the included Windows Template.

One system shows Disk Usage, Memory (Physical and Virtual), CPU activity, Logged in users, Processes, and Ethernet activity. The other system shows all of the same items except the memory info.

Is this normal? Is there something I need to do to get more consistent results?

I actually have 3 different Cacti systems running 3 different versions and all of them have this same inconsistency.

Any help would be appreciated.

Thanks in advance,

Jared

[BSOD2600]

In the SNMP service, check if all the Services are enabled under the Agent tab.

[jrdbeau]

In the SNMP service, check if all the Services are enabled under the Agent tab.

Yep, everything is checked on both systems. Also, for the system that is not showing the memory info, Cacti seems to lose contact with that system from time to time. It shows "Down" in the devices list and "SNMP error" in the device detail page. Could it be a problem on both ends? It just seems odd since both target systems are identical in every way.

Jared

[lard]

Hi,

If you are also getting missed SNMP Polls then I would increase the timeout for that one host and see if this sort it - FYI we have a very busy IPT server than more often than not ignores SNMP queries when it decides that another process takes priority, noted the same in routers in that if they are busy the first process to get bumped is SNMP - is one of these two servers more utilisied than the other?

One other thing to check is that you can see the memory info via perfmon on the server - if SNMP informant uses the perfmon database to supply the polled info then it may be this that is at fault,

HTH,

Lard

[jrdbeau]

Hi,

If you are also getting missed SNMP Polls then I would increase the timeout for that one host and see if this sort it - FYI we have a very busy IPT server than more often than not ignores SNMP queries when it decides that another process takes priority, noted the same in routers in that if they are busy the first process to get bumped is SNMP - is one of these two servers more utilisied than the other?

One other thing to check is that you can see the memory info via perfmon on the server - if SNMP informant uses the perfmon database to supply the polled info then it may be this that is at fault,

HTH,

Lard

No dice. Both systems have very low utilization right now and I also verified that the perfmon memory settings are the same. The strange thing is that on a different Cacti server that I have running, both machines report correctly. It seems random.

Thanks for the suggestions...

Jared

[lard]

Hi,

If you are also getting missed SNMP Polls then I would increase the timeout for that one host and see if this sort it - FYI we have a very busy IPT server than more often than not ignores SNMP queries when it decides that another process takes priority, noted the same in routers in that if they are busy the first process to get bumped is SNMP - is one of these two servers more utilisied than the other?

One other thing to check is that you can see the memory info via perfmon on the server - if SNMP informant uses the perfmon database to supply the polled info then it may be this that is at fault,

HTH,

Lard

No dice. Both systems have very low utilization right now and I also verified that the perfmon memory settings are the same. The strange thing is that on a different Cacti server that I have running, both machines report correctly. It seems random.

Thanks for the suggestions...

Jared

No Problem - suppose the next suggestion would be to run a poll in debug and see what the output is (not that I would know what I was looking for )

Either that or can you run a straight SNMP walk against the oid and see what is returned? If the returned output is the same for both servers you could at least pin it to Cacti and then run the debug?

Lard