When adding new devices to my Cacti instance, I get frequent "SNMP error" messages in the device screen. But the error is not consistent, not even for the same device. Here's what I already have checked:

• Sometimes a device shows that "SNMP error" message even when it did not had that error an hour before, and vice versa.
• I tried this with several different Cacti releases, installed on different OS (Debian squeeze: 0.8.7g-1+squeeze1, Debian Sid: 0.8.7i-3, CentOS 6.0: 0.8.7i-2.el6)
• tried both from a local (192.168.1.xy) network and from a different data center so I don't think it is a network problem
• reinstalled the Cacti database, rerun the scripts to install my devices. Now different devices have that error
• when executing a snmpwalk or snmpgetnext command from the command line, it is always successful
• increasing the timeout to 20000 (20 seconds) and the retry count to 10 does not make a difference

The cacti.log says:

04/14/2012 02:10:19 PM - CMDPHP: Poller[0] WARNING: SNMP GetNext Timeout for Host:'s0026.mydomain.de', and OID:'.1.3.6.1.2.1.1.3.0'
04/14/2012 02:10:20 PM - CMDPHP: Poller[0] WARNING: SNMP GetNext Timeout for Host:'s0026.mydomain.de', and OID:'.1.3'

However, when executing snmpget or snmpget with that from the command line a proper response is returned immediately.