No problem for the delay. We did 2 things to stabilize the environnement. First we moved the zabbix server VM to a less busy physical server. Second for particular reasons we had to have the zabbix server answering on 2 differents TCP ports. The first TCP port was the regular 10051 which was handled directly by the zabbix_server process, and we also had the 10053 port handled by xinetd and which was forwarded to 10051 port. We worked at our setup to simplify things and managed to remove the tcp forwarding done by xinetd. So far the crash occurences seemed to disappear. We let 1.8.10 in place.