Want your servers monitored without maintaining your own monitoring platform? Server monitoring as a service starts at 49.90 euros net per server per month. For your own Zabbix platform there is Managed Zabbix.
Server monitoring reports problems before users notice them: a file system filling up, a crashed service, an expiring certificate, a backup that has been failing for days. The German BSI IT-Grundschutz lists as a standard requirement ('should') in SYS.1.1.A23 that the system state and services of a server are monitored continuously and that threshold violations are reported to operations staff (BSI SYS.1.1, edition 2023, German).
This guide shows which values should be monitored, which thresholds the common templates use, which tools are available under which license, and when running it yourself or outsourcing makes sense. The cost question in detail is answered in What does server monitoring cost?.
Contents
- What server monitoring does
- Checklist: what needs to be monitored
- Collection methods: agent, pull, push and outside view
- Server monitoring tools compared
- Open source or commercial
- Alerting that works in operations
- Run it yourself or outsource
- Our approach at WZ-IT
- Status and methodology
What server monitoring does
Monitoring answers three questions: Is the system running? Is it running well? Will it still be running tomorrow? Availability checks answer the first, performance data the second, trends the third, for example the growth of a file system over weeks.
Monitoring detects and reports. It fixes nothing. An alert without an owner does not change the state of a server. Every monitoring concept therefore needs three decisions: which values are measured, at which value an alert is raised, and who handles it within what time.
Checklist: what needs to be monitored
The table summarises the baseline for a Linux server. The thresholds come from the official Zabbix template "Linux by Zabbix agent" and the certificate template for Zabbix agent 2 (Linux template, certificate template). They are starting values and are adjusted per system.
| Area | Metrics | Default value in the Zabbix template | Why |
|---|---|---|---|
| Availability | ping, agent reachable, SSH port | agent unavailable for 3 minutes | First sign of an outage or network problem |
| CPU and load | utilisation, load per CPU | CPU above 90% for 5 minutes; load per CPU above 1.5 for 5 minutes | Bottlenecks and hanging processes |
| Memory | utilisation, swap | memory above 90% for 5 minutes; free swap below 50% | Precursor to OOM kills and crashes |
| File systems | usage, inodes | warning from 80%, critical (severity Average) from 90%; free inodes below 20% or 10% | Full file systems stop databases and logging |
| Disk I/O | read and write latency | wait time above 20 ms for 15 minutes | Slow disks slow down every service |
| Network | utilisation, errors | utilisation above 90% as a 15-minute average, error packet rate above 2 for 5 minutes | Overloaded or faulty interfaces |
| Services | processes, ports, HTTP response | per service | A running server with a stopped web server is an outage |
| Certificates | TLS expiry date | warning 7 days before expiry | Expired certificates take down websites and APIs |
| Backups | success, age of the last backup | per backup tool | Otherwise a failed backup is only noticed at restore time |
| Updates | pending security updates, reboot required | per distribution | Otherwise open vulnerabilities go unnoticed |
| Hardware | SMART values, RAID status, temperature | per vendor | Only on bare metal and own hardware |
Applications need their own checks: response times, database connections, queues, error rates in logs. With containers and hypervisors, host and guests are monitored separately, for Proxmox for example with the official template (Monitor Proxmox with Zabbix).
Collection methods: agent, pull, push and outside view
- Agent-based: a small service on the server delivers metrics, including those not visible from outside, such as file systems or processes. Zabbix, Checkmk and Icinga work this way.
- Agentless: queries via SNMP, SSH or APIs. Common for network devices and appliances where no software can be installed.
- Pull: the monitoring server queries its targets at fixed intervals. Prometheus collects time series through an HTTP pull model from exporters such as the Node Exporter (Prometheus).
- Push: the target reports itself, for example a backup job that calls a URL after a successful run. If the report is missing, an alert is raised.
- Outside view (uptime monitoring): checks from outside your own network whether a website or service is reachable and responds correctly. Only the outside view detects DNS, routing and firewall problems the way users experience them.
A complete setup combines inside and outside view. The monitoring itself should not run on the same infrastructure as the monitored systems, otherwise it fails together with them when it matters.
Server monitoring tools compared
| Tool | License | Cost | Strength | Suited for |
|---|---|---|---|---|
| Zabbix | AGPLv3 since version 7.0 | all features free; optional vendor support | Integrated platform with agent, templates, escalations, dashboards | Servers, VMs, network, mixed environments |
| Checkmk | Community Edition GPLv2; Pro, Ultimate, Cloud commercial | Community 0 euros; Pro from 190 euros, Ultimate from 275 euros per month billed annually; Cloud from 240 euros per month (1,000 services) billed monthly | Automatic discovery, quick start | Environments with many standard systems; Community up to about 100 hosts according to the vendor |
| Prometheus and Grafana | Prometheus Apache 2.0, Grafana AGPLv3 | free; effort for exporters, rules, storage | Time series, containers and Kubernetes, flexible queries | Cloud-native environments with a platform team |
| Icinga 2 | GPLv3 or later since version 2.16.0 | free; optional vendor support | Nagios-compatible checks, distributed setups | Environments with existing Nagios checks |
| PRTG | proprietary | freeware up to 100 sensors; subscriptions by sensor count, PRTG 500 from 200 US dollars per month billed annually according to Paessler | Many ready-made sensor types, Windows server as base | Windows-heavy environments, network |
| Uptime Kuma | MIT | free | Outside view for HTTP(S), TCP, ping, DNS, Docker; status pages; 90+ notification services | Reachability of websites and services, no resource metrics such as CPU, memory or file systems |
Sources: license files of the projects on GitHub (Zabbix, Checkmk, Prometheus, Grafana, Icinga 2, Uptime Kuma), Checkmk pricing, PRTG pricing, PRTG licenses.
Zabbix covers data collection, alert logic, escalation chains and dashboards in one platform. There is no commercial edition with additional features. The direct comparisons are in Zabbix vs. Checkmk and Zabbix vs. Prometheus.
Checkmk discovers services on a host largely automatically. The commercial editions are billed by monitored hosts or services, so costs grow with the environment.
Prometheus stores metrics as time series and is the standard in Kubernetes environments. Visualisation is usually done with Grafana, alerting with Alertmanager. For classic servers the setup is more involved than with Zabbix or Checkmk, because exporters, rules and storage are configured individually.
PRTG counts sensors: one sensor monitors a single value. How the sensor model adds up is shown in Zabbix vs. PRTG.
Uptime Kuma does not replace server monitoring but adds the outside view. Run on a separate system, it still reports when the monitored infrastructure fails completely.
Open source or commercial
Open source tools remove the license fee, not the effort. The platform needs a server, a database, updates, backups and maintenance of templates and thresholds. Commercial products often bundle convenience features and vendor support, in exchange for recurring fees based on size.
Three questions help with the choice:
- What is being monitored? Classic servers and network point to Zabbix, Checkmk or Icinga, containers and Kubernetes to Prometheus.
- Who maintains the platform? Without an owner, templates and thresholds go stale and monitoring reports too much or too little.
- How will the environment grow? With licenses per host, service or sensor, costs rise with every new system; with free tools only the operating effort grows.
Alerting that works in operations
- Every alert needs an action. Alerts nobody has to act on belong in reports, not notifications. Otherwise the team gets used to noise and misses the important ones.
- Separate severities: a warning at 80% usage can wait until the next business day, a failed service cannot.
- Define escalation: who is notified first, who after how long, through which channel?
- Register maintenance windows: planned reboots should not trigger alerts.
- Monitor the monitoring: a second, independent system or an external check reports when the monitoring platform itself fails.
Run it yourself or outsource
| Run it yourself | Monitoring as a service | Management including monitoring | |
|---|---|---|---|
| Platform | own installation, own maintenance | provider runs it | provider runs it |
| Review alerts | own team | provider reviews and notifies | provider reviews and acts |
| Resolution | own team | own team | provider within the agreed scope |
| On-call at night | own on-call rota needed | depending on contract | depending on service level |
| Suited for | teams with admin capacity and on-call | teams that resolve issues themselves but do not want to monitor | companies without their own Linux operations |
What matters is not the tool but the chain from measurement to resolution. The checklist for choosing a provider is in Outsourcing IT monitoring.
Our approach at WZ-IT
Monitoring Only starts at 49.90 euros net per server per month. Monitoring runs 24/7 and covers system metrics, services and reachability, SMART values on own hardware as well as certificates and simple HTTP checks. Our team reviews every alert and notifies the agreed contact during service hours; resolution remains with the customer.
If you want your own monitoring platform, Managed Zabbix covers setup, integration and ongoing operation of a Zabbix environment by scope. If WZ-IT should also respond and resolve, Linux server management from 149.90 euros net per server per month includes monitoring, updates, backup verification and incident response. Response to P1 incidents follows the booked service level. We manage servers remotely, at the existing provider or on-premises.
Further guides
- What does server monitoring cost?: license, operations and staff compared
- Outsourcing IT monitoring: checklist for choosing a provider
- Zabbix vs. Checkmk, Zabbix vs. Prometheus, Zabbix vs. PRTG
- Linux server maintenance checklist: monitoring as part of ongoing maintenance
Status and methodology
- Collected: 26 October 2026.
- Thresholds: macros of the official Zabbix templates on the master branch of the Zabbix repository. Other tools use their own defaults.
- Licenses: license files of the project repositories, for Icinga 2 additionally the license statement in the README. Checkmk and PRTG according to vendor pages.
- Prices: vendor list prices on the collection date, Checkmk in euros, PRTG in US dollars.
- Not included: SaaS services for uptime monitoring, log management and APM tools.
Sources
- BSI IT-Grundschutz SYS.1.1 General server (edition 2023, German)
- Zabbix template "Linux by Zabbix agent"
- Zabbix template "Website certificate by Zabbix agent 2"
- Checkmk pricing
- PRTG pricing and PRTG manual: available licenses
- Prometheus on GitHub
- Grafana on GitHub
- Icinga 2 on GitHub
- Uptime Kuma on GitHub
