As to http://ide-x.org/openhub/zabbix/blob/master/zabbix-2.4.0/src/zabbix_server/watchdog/watchdog.c, it might be too sensitive for each time failed to connect, then send out email alert.

Is it possible to enhance monitor strategy to minimize sudden alert due to network fluctuation?
Within time period, collect some fault status, then determine it's trustable status ot nor based on given trust threshold. Like Cassandra Phi fault detection: ddg.jaist.ac.jp/pub/HDY+04.pdf