Guide and tool
Server monitoring: what to measure and how to get it running in minutes
Server monitoring means continuously measuring load and availability and alerting automatically when something deviates, before users notice the fault. It takes two perspectives: an agent on the server for CPU, memory and disks, and a check from outside for what visitors actually experience.
As of 5 November 2026
Notifications
Metrics and thresholds
Which values should you monitor on a server?
CPU, memory, disk space, running services and whether the server is still reporting at all. The table shows what each value tells you and when an alert makes sense.
| Value | Starting point for the alert | What it tells you |
|---|---|---|
| CPU usage | above 90% for 5 minutes Preset in inping | Is the server working at its limit? Short spikes are normal, sustained load is not. |
| Memory | above 90% for 5 minutes Preset in inping | Memory in use, not counting the file cache. When it fills up, the system ends processes or starts swapping. |
| Disks | above 90%, immediately Preset in inping | Disks fill steadily and do not recover by themselves. A full disk stops databases and logs. |
| Docker services | not running, immediately Preset in inping | A container that has stopped or is stuck in a restart loop. |
| Agent offline | 3 missed reports, so 3 minutes at a 60-second interval Preset in inping | The server has stopped reporting: a crash, the network, a stopped service. |
| Load | your own rule, for example load above the core count for 15 minutes Our suggestion, as your own rule in inping | Processes that want to run or are waiting. Staying above the number of CPU cores means work is piling up. |
| Swap | your own rule, for example above 50% for 15 minutes Our suggestion, as your own rule in inping | Constant swapping slows the whole server down and announces a memory shortage. |
Why 5 minutes for the CPU but immediately for the disk?
A CPU may sit at 100% for a moment: a build or a backup is not an incident. Only sustained load is a problem, so the rule applies only when every value in the time window is above the limit. A disk, on the other hand, moves slowly and in one direction. When it reaches 90%, it will still be full five minutes later, so the alert goes out immediately.
When should you depart from the starting values?
Watch a normal week, then set the limits. 10% free is 200 GB on a 2 TB disk, but only 2 GB on a 20 GB disk: there, a rule in absolute bytes is better. A database server keeps a lot in memory on purpose and can sit higher than a web server. A rule in inping always reads the same way: when a value is above, below or equal to a limit, for a number of minutes.
Two perspectives
Agent or external check: which is better?
Neither on its own. The external check answers “Does it work?”, the agent answers “Why not?”. A server that has failed cannot report anything, and a website can be unreachable although its server is healthy.
Three cases only the combination can explain
- Website down, agent reports normal values: DNS, certificate, web server configuration or the route to it. The error type of the check narrows it down.
- Website slow, memory at 97%: Memory is the likely cause. The agent shows which Docker service or which account is using it.
- Website down, agent offline: The server or its network has failed. inping reports both as separate events.
Several runners do not count an outage twice: an incident opens when the platform's verdict across all runners is down, and closes when it is up again.
Approaches compared
Five ways to monitor a server
Each approach solves a different problem. inping combines the external check and the agent as a hosted service.
| Approach | What it sees | Effort | Limit |
|---|---|---|---|
| External HTTP check | Availability, status code, response time and the kind of error, such as DNS, TLS or a timeout: what visitors experience. | Nothing on the server, you only enter the URL. | Tells you that something is broken, not why. |
| Agent on the server | CPU, memory, disks, services, installed packages, scheduled jobs. | Install once, then updates. | A server that has failed reports nothing. You need an offline alert. |
| Agentless via SNMP, WMI or SSH | Devices that cannot run software, such as switches and printers. | No installation, but credentials and open ports. | Usually less depth on applications. inping does not offer this approach. |
| Self-hosted suite | Almost everything, for example with Zabbix, Icinga or Prometheus and Grafana. | Your own server; upkeep, updates and alert paths are yours. | If the monitoring goes down with the monitored network, it stays silent. |
| Hosted service | Depends on the provider. inping combines external checks and an agent. | Create an account, ready in minutes. | Data sits with the provider, so check its location and the data processing terms. |
We compare specific products in our comparison of monitoring tools.
What to look for in a server monitoring tool
Both perspectives
A check from outside and metrics from inside. Only together do they show whether a website is down and why.
Alerts that close
An alert should open once and close once. Anyone who gets a mail for every measurement soon stops reading them.
Independent of what it watches
Monitoring must not depend on the same machine or network as the monitored service, or it goes quiet when it matters.
Least privilege
The agent only needs read access and outbound HTTPS. If a tool asks for root or open ports, ask why.
Clear data location
Logs and hostnames can be personal data. The provider should disclose its location, deletion periods and data processing terms.
Setting up inping
How do I set up monitoring in a few minutes?
Four steps, with no open port and no need for root. The example shown is for Linux with systemd.
1. Create an account and a project
Sign up and create one project per application. Enter the URLs that must be reachable. The checks run from then on.
2. Install the agent
Under Agents → Add agent inping shows the finished command for your system, with a token that is valid for 24 hours and shown only once. Here it is with placeholders:
curl -fsSL https://www.inping.io/install.sh | INPING_TOKEN=<token> sudo --preserve-env=INPING_TOKEN sh -s -- --project <project-id>=/srv/shopThe script verifies the SHA-256 sum of the download, creates the user
inping-agentand starts the service. Guides: Monitor a Linux server, Monitor a Windows server, Monitor Docker containers, and all systems in the installation guide.3. Review the alert rules
Every new server starts with the four preset rules from the table above and the offline alert. You add your own rules on the agent's page.
4. Decide on notifications
Rules per project send events by mail to the team, to up to ten external addresses or to your clients' contacts, or as a signed webhook to your system. External addresses confirm by double opt-in first.
The result
What you see afterwards
All tiles reproduce the interface, here with example data.
Vulnerabilities in shop.example.com
Beyond load, the agent reads installed packages and runtimes and, through the platform, compares them daily with known vulnerabilities, lists scheduled jobs and can send server logs if you want it to. More under Features and in the public security advisories.
To be honest
What inping is not the right tool for
So that you do not find out only after installing it.
- No ping, port or SNMP checks. Uptime is measured over HTTP(S). For network devices without their own agent, see what we describe in Monitor UniFi and Monitor a Synology NAS.
- No promised availability: inping currently offers no SLA, and the terms of use exclude any guarantee.
- No legal guarantee. What stays with you on data protection is explained under GDPR server monitoring.
Frequently asked
Questions about server monitoring
What is the easiest way to monitor a server?
Combine an external HTTP check of your most important URLs with an agent on the server, and decide who is notified about which event. With inping that takes an account, a project with URLs and one install command.
Which thresholds make sense for server monitoring?
As starting values, inping sets these on every new server: CPU above 90% for 5 minutes, memory above 90% for 5 minutes, any disk above 90% and any service that is not running. They are starting values, not a standard: after a week of operation you can see what is normal for you and adjust the rules.
Can I monitor a server for free?
Yes. The Free plan of inping costs nothing and needs no payment details: 1 project, 2 agents, 1 own runner and 10 URLs per project, with a check interval from 60 seconds.
How often should a server be measured?
For most servers one measurement per minute is enough, which is the default of the inping agent. You can set anything from 10 seconds to one hour. Raw values stay for 7 days, hourly averages, minimums and maximums for 365 days.
Is a monitoring agent a security risk?
Any additional software is attack surface, so the way it is built matters. The inping agent runs under its own user, writes only to its state directory, opens no port and makes outbound HTTPS connections only. Updates are signed and verified before they are installed.
What happens when the server cannot reach the platform?
The agent keeps measuring, buffers the data for up to 24 hours by default (64 MB at most) and sends it as soon as the connection is back. The platform discards duplicates.
Can inping monitor ping, ports or SNMP?
No. The uptime checks of inping are HTTP(S) requests, and server values come from the agent. inping does not offer ping, plain port checks or SNMP. For pure network devices, a tool with SNMP is the better choice.
Which systems does the agent run on?
On Linux, macOS, Windows and FreeBSD (amd64 and arm64), and as a Docker image in a Compose file. On Linux and macOS it can also be installed without root, in your own user account.
Your first server is connected in minutes
Free and without payment details. The plans are listed under Pricing.