While querying SNMP-capable devices is a nice method that requires little or no configuration of each device in itself, in some situations, information flow in the reverse direction is desired. For SNMP, these are called traps. Usually, traps are sent upon some condition change, and the agent connects to the server or management station on port 162 (as opposed to port 161 on the agent side, which is used for queries). You can think of SNMP traps as being similar to Zabbix active items; as with those, all connections are made from monitored machines to the monitoring server.
The direction of the connections isn't the only difference—SNMP traps have some other pros and cons when compared to queries. For example, SNMP traps are usually more capable of detecting short-lived problems that might have been missed by queries. Let's say you are monitoring incoming voltages on a UPS. You have decided on a reasonable item interval that would give you useful data and wouldn't overload...