Network and server monitoring that warns you before an outage

Network & Server Monitoring: a practical guide

Zabbix or PRTG for network monitoring?

PRTG is commercial software licensed by sensor count, and its core server runs only on Windows. Initial setup is quick and its auto-discovery is good, but for Iranian companies, buying and renewing a valid license is difficult because of sanctions.

Zabbix is open source with no licensing limits, runs on Linux and extends to branches and separate networks with Zabbix Proxy. The trade-off is more time spent on templates and thresholds. For multi-branch networks or infrastructure with many monitored items, Zabbix is usually the more practical choice.

Linux server monitoring: items that matter more than CPU

Running out of inodes stops a server from creating files even with free disk space, and default disk checks do not show it. The timeleft function in Zabbix predicts when a disk will fill based on its growth trend, which is more useful than a fixed percentage alert.

Clock drift against NTP breaks Active Directory authentication and certificate validation. Software RAID status from /proc/mdstat, disk health via smartctl, failed systemd units and the time of the last successful backup are not in the default templates and need to be added. Always read load relative to the number of cores.

Server room monitoring: temperature and UPS

If the server room air conditioner fails at night or over a weekend, servers and disks can be down by morning. Inlet temperature can be read from iLO on HPE servers or iDRAC on Dell over IPMI or Redfish, and a network temperature sensor with SNMP support covers the room itself.

UPS units with a network card are polled over SNMP, and for the rest NUT reads data over USB. An on-battery alert with the remaining runtime gives you time to shut servers down cleanly before the battery runs out.

Too many alerts: dependencies and maintenance windows

When a branch router goes down, every device behind it looks down too and dozens of alerts fire at once. With trigger dependencies in Zabbix, only the router alert goes out and the rest stay quiet until it recovers.

Define maintenance periods for planned work such as updates and reboots so they do not raise false alarms. Route alerts by severity: Disaster goes by SMS to whoever is on call, Warning only to email and the daily summary. If nobody acknowledges an alert within a set time, escalation passes it to the next person.