Description

How to minimize SNMP load to routers

Symptoms

When SNMPD processes a lot of incoming requests, the Unix domain socket connecting snmpd to a subagent could get full.  This usually happens when the subagent is slow in retrieving data from the kernel/PFE or it's backed up for other reasons. The following messages could be seen in log messages:

Apr 1 08:20:31 router mib2d[3657]: %DAEMON-3-SNMP_IPC_ERROR: Header version mismatch for master/subagent message: sock:16 (type:1)
Apr 1 08:20:31 router mib2d[3657]: %DAEMON-3-SNMP_SMS_HDR_ERR: problem with hdr size (6) or msg size (0)
Apr 1 08:20:32 router mib2d[3657]: %DAEMON-3: snmp_ipc_try_connect: opened socket 16

Solution

To minimize possibilities of running into this, the following "best practice" should be exercised:
 

  • Poll devices by row rather than by column
  • Pack fewer OIDs into a single packet
  • Configure set snmp filter-duplicates
  • Reduce the polling interval and/or number of polling stations per router
  • Increase snmp request time out value to reduce snmpd load. Do not use sub-second timeout