How to Monitor VPS Performance
By Harborvm7 min read
Key takeaways
- Watch four metrics on a Linux VPS: CPU load, memory use, disk I/O and network throughput.
- Start with built-in tools (top, htop, vmstat, iostat) before installing anything.
- A lightweight agent like Netdata gives you history and graphs without a heavy footprint.
- Set alerts on thresholds you choose so you hear about problems before users do.
- Read trends over weeks, not single spikes, to decide when to resize your plan.
To monitor VPS performance, track four things over time: CPU load, memory use, disk I/O and network throughput. Start with built-in Linux tools for a live look, then add a lightweight agent that keeps history and sends alerts when a metric crosses a line you set.
Good monitoring is less about fancy dashboards and more about knowing what normal looks like for your server. Once you know that, a spike or a slow climb tells you something is wrong, and you can act before your site goes down or your database starts swapping. This guide walks through the metrics that matter, the tools already on your VPS, how to add a small agent, how to set alerts, and how to read trends so you resize at the right time.
The stakes are real for anyone running a service that people depend on. Research from Splunk and Cisco put the average cost of downtime at $15,000 per minute for large enterprises. Your VPS is smaller, but the lesson holds: catching a problem early is far cheaper than cleaning up after an outage.
The key metrics: CPU, RAM, disk I/O and network
Four numbers tell you most of what you need to know. Learn what each one means and what "too high" looks like.
CPU. This is how busy your processor is. On Linux you often see it as load average, three numbers for the last 1, 5 and 15 minutes. A rough rule: if the load average stays above the number of CPU threads you have for long stretches, work is queuing up. On a Harborvm VPS, 1 vCPU equals one real CPU thread, so a Small plan (1 vCPU) that sits at a load of 2 or 3 is overloaded. Watch for sustained high usage, not brief spikes during a deploy or a backup.
RAM. Memory fills with your running programs plus the kernel's disk cache. Free memory near zero is not itself a problem, because Linux uses spare RAM for cache. The warning sign is swap. When the server runs out of real memory and starts moving pages to disk, everything slows down. If you see steady swap activity, you need more memory or a lighter workload.
Disk I/O. This is read and write activity on your storage, measured in operations per second and throughput. High I/O wait means your CPU is sitting idle waiting for the disk. Databases and busy logs are the usual causes. Harborvm uses SSD storage, which handles random reads and writes far better than spinning disks, but heavy write loads can still saturate it.
Network. This is traffic in and out, measured in megabits per second, plus packet counts and errors. Watch for a link that is pinned at its limit during normal hours, and watch your monthly transfer so you do not get surprised.
Built-in tools: top, htop and vmstat
Before you install anything, your VPS already ships with tools that give a live picture. These are the first things to reach for when a server feels slow.
top is on every Linux system. Run it:
top
The header shows load average, CPU breakdown and memory. The list below shows processes sorted by CPU use. Press M to sort by memory instead, and q to quit. It is plain and always available.
htop is friendlier. It adds colour, per-core bars and easy scrolling. Install it if it is not already there:
# Ubuntu or Debian
sudo apt install htop
# AlmaLinux, Rocky or Fedora
sudo dnf install htop
Then run htop. You get a clear view of each CPU thread, a memory bar and a sortable process list. You can scroll sideways to see full commands and kill a runaway process with F9.
vmstat shows system-wide activity in one line per interval. Run it with a two-second refresh:
vmstat 2
Watch the si and so columns for swap in and swap out; anything above zero for long means memory pressure. The wa column under CPU is I/O wait; a high number points at the disk.
For storage specifically, iostat (from the sysstat package) breaks down per-device read and write rates:
iostat -x 2
Look at %util for how busy a device is and await for how long requests wait. These four tools answer most "why is it slow right now?" questions without any setup.
Setting up a lightweight monitoring agent
Live tools are great in the moment, but they do not keep history. You cannot answer "was the server busy at 3 a.m.?" with top. For that you need an agent that records metrics over time and draws graphs.
Harborvm already does part of this for you. Every VPS shows live CPU, memory, disk and network graphs in the control panel, so you can check your server from any browser without logging in over SSH. That covers the common case with nothing to install.
When you want deeper, per-process history on the machine itself, add a small agent. Netdata is a good fit for a single VPS: it collects thousands of metrics per second, stores recent history and serves its own dashboard, all with a light footprint. Install it with the official script:
wget -O /tmp/netdata-kickstart.sh https://my-netdata.io/kickstart.sh
sh /tmp/netdata-kickstart.sh
By default it listens on port 19999. On Harborvm you do not get a dedicated public IPv4 address, so reach the dashboard one of two safe ways: forward a port to it, or better, connect over your private WireGuard VPN so the dashboard never faces the open internet.
If you run several servers and want one place to see them all, Prometheus with Grafana is the common choice. It is heavier to set up, but it scales. It is also widely used: in Grafana Labs' 2024 observability survey, 75% of organisations reported running Prometheus in production. For a single VPS, Netdata is simpler; reach for Prometheus when you outgrow one box.
Configuring alerts before problems escalate
Graphs only help if you are looking at them. Alerts do the watching for you and tell you when a metric crosses a threshold you set.
Pick thresholds based on what you learned about normal load. Sensible starting points:
- CPU load average above your vCPU count for more than 5 minutes.
- Memory use that pushes the server into steady swapping.
- Disk usage above 85 per cent, so you fix it before the disk fills.
- Any sustained I/O wait that correlates with slow response times.
Netdata ships with sensible default alerts and can send notifications to email, a webhook or chat tools. You configure recipients in its health_alarm_notify.conf file. If you prefer to keep it simple, a short shell script run by cron can check a value and email you when it crosses a line. The point is to be told, not to remember to look.
One honest limit to plan around: Harborvm does not offer backups or snapshots. Alerts help you react fast, but they are not a safety net. Keep your own copies of anything you cannot lose, and test that you can restore them.
Interpreting trends to plan upgrades
The real value of history is spotting the slow climb. A single spike at deploy time is noise. A metric that creeps up week after week during normal hours is a signal.
Here is how to read each one:
- CPU trending up. If your load average sits higher each month at the same traffic, your app is doing more work per request, or you have more traffic. Add vCPUs when the load regularly meets or exceeds your thread count.
- Memory trending up. Rising baseline memory with occasional swap means you are close to the edge. More RAM is usually the cheapest fix and removes the biggest source of slowness.
- Disk filling. Watch the slope, not just the current number. If you add 1 per cent a day, you know roughly when you will hit the wall. Clear old logs first, then grow the disk if the data is genuinely needed.
- Network near the limit. If your link or transfer sits near its ceiling during normal hours, size up before growth forces the issue.
When one resource is the constraint and the others have room, a custom size can be cheaper than jumping a whole tier. Harborvm plans run from Small (1 vCPU, 2 GB, 40 GB) up to X-Large, with custom sizes to 16 vCPU, 64 GB and 500 GB, so you can add just the resource the trend points at. For the reasoning behind dedicated threads, see what a dedicated vCPU is and why it matters.
The next step is practical: open a terminal on your VPS, run htop and vmstat 2 for a few minutes during a busy period, and write down what normal looks like. Those numbers are your baseline, and everything else, agents, alerts and upgrades, builds on them.
Frequently asked questions
What does it mean to monitor VPS performance?
It means tracking how your server uses CPU, memory, disk and network over time, so you can spot bottlenecks, fix problems early and size your plan correctly.
Which tool should I use to monitor a Linux VPS?
Start with built-in tools like htop, vmstat and iostat for live checks. For history and alerts, add a lightweight agent such as Netdata.
How do I know when to upgrade my VPS?
Look at trends over weeks. If a metric stays near its limit during normal load and keeps climbing, resize that resource rather than waiting for an outage.
Does Harborvm include monitoring?
Yes. Every Harborvm VPS shows live CPU, memory, disk and network graphs in the control panel. You can also run your own agent inside the VM.