Most of the outages we fix had given warning. The disk had been filling up for three weeks. The certificate showed its expiry date. The backup failed every night with the same message. Nobody was looking. Monitoring is simply deciding to look — automatically, continuously, and to be told before the user calls.

What to watch first

Twelve indicators cover the essentials for an SMB:

  1. Disk space on every server, with a threshold at 80 %.
  2. Backup status: did the last one succeed?
  3. Service availability: website, email, business application, from outside.
  4. SSL certificate expiry, 30 days ahead.
  5. Domain name and licence expiry.
  6. CPU load and memory abnormally high.
  7. Disk temperature and health (SMART, degraded RAID).
  8. UPS: on battery, battery worn.
  9. Network links: main internet link down, failover to backup.
  10. Security updates pending for too long.
  11. Failed login attempts in bulk.
  12. Response time of the business application, to see degradation before failure.

Which tool to choose

PRTG: the simplest to deploy, sensor-based, with a clear interface. Free up to 100 sensors, which covers a small infrastructure. A good starting point.

Zabbix: open source, very complete, auto-discovery, templates for everything. Needs some learning time, but no licence cost.

Centreon: open source of French origin, designed for operations teams, with mapping and availability reports.

Grafana + Prometheus: the reference for containerised and cloud environments, metrics- and dashboard-oriented.

Dynatrace: advanced application observability with automatic root-cause analysis, for critical, high-load applications. Oversized for a classic SMB.

For a Tunisian SMB with a few servers, we most often start with Zabbix or PRTG, and add Grafana when there are applications to follow closely.

The real subject: alerts

Monitoring that sends 50 emails a day is ignored within a week. Three rules:

  • One alert = one action. If nobody has to do anything on receiving it, it is not an alert, it is a chart.
  • Thresholds that rise in steps: warning at 80 %, critical at 95 %, with different channels (email, then SMS or Telegram).
  • A monthly review of received alerts to remove those that served no purpose.

And at night?

Monitoring detects; someone still has to react. For an SMB, three levels exist: business hours only, internal on-call, or on-call entrusted to a provider. The right level depends on your applications' RTO, which we explain in RTO and RPO: how long can you stop?.

Semocode deploys and operates Zabbix, Centreon, PRTG and Grafana monitoring, with a prior audit of what is not covered today. See Support & Maintenance or book a call.