I have been struggling with getting SNMP low level discovery working correctly on a variety of Cisco devices I have at work (ranging from Cisco 2811 and 3825 routers, to 6509 switches, to 4948 switches). I have opened and added to other tickets while banging my head against this problem, thinking originally that it was an SNMP timeout issue (zabbix_server.log at Debug=4 seemed to indicate it was a timeout issue, at first)

I just changed two of my Cisco devices over from being SNMPv3 to being SNMPv2, and it looks like they're suddenly polling correctly... the low level discovery items all create perfectly fine, and bandwidth graphs based on ifHCInOctets and ifHCOutOctets are rolling in.

The big difference between using SNMPv3 and rolling back to SNMPv2 as I can see it in zabbix_server.log is that the " SNMP item [ifOutErrors-LAN] on host [cisco6509] failed: first network error, wait for 3 seconds" have been replaced with "became not supported: No Such Instance currently exists at this OID" events

This is leading me to believe that the "No Such Instance currently exists at this OID" handling for SNMPv3 item checks has a bug... instead of interpreting the SNMPv3 equivalent of "No Such Instance currently exists at this OID" correctly and marking a specific item not supported, the zabbix snmp poller seems to be interpreting the SNMPv3 version of that message as a network error, and enough of these causes the whole SNMP host to go into an error state. This causes SNMPv3 hosts that I have set up with LLD to stop polling essentially, because as soon as item polling starts, zabbix interprets errors and marks the parent host as down, instead of interpreting the responses correctly and marking items as 'not supported.'