Published: 23 September 2026
A load test is the textbook case for renting compute by the hour. You need several machines, in more than one place, for roughly one afternoon, twice a year. Buying that capacity monthly is absurd; buying it for four hours costs less than lunch.
Why you cannot load-test from your laptop
Three reasons, all of which produce numbers that look fine and mean nothing.
First, your home or office uplink saturates long before your server does. You measure your ISP, not your application. Second, a laptop cannot hold tens of thousands of open sockets — you run out of ephemeral ports and file descriptors and start measuring your own client. Third, the path from your desk to your server is nothing like the path your users take, so the latency distribution you record is fiction.
A generator sitting in a datacentre, on a 1 Gbps port, in a country where your users actually live, fixes all three.
Sizing a generator
Load generators are bound by CPU and by open connections, not by disk. The rough shape, against our hourly VPS plans:
| Plan | RAM / vCPU | Reasonable target | Per hour (Bucharest) | Per hour (Frankfurt) |
|---|---|---|---|---|
| VPS II KVM | 4 GB / 2 | A few thousand virtual users on a simple HTTP scenario | €0.0083 | €0.0125 |
| VPS III KVM | 8 GB / 2 | Scenario scripts with think time and parsing | €0.0139 | €0.0222 |
| VPS IV KVM | 16 GB / 4 | The workhorse generator — high connection counts, TLS handshakes | €0.0222 | €0.0361 |
| VPS VI KVM | 32 GB / 8 | One big generator instead of four small ones | €0.0806 | €0.1084 |
Treat those as starting points, not promises — the real number depends entirely on how heavy your scenario script is. The honest method is to ramp one generator until its own CPU is the limiting factor, note the rate, and scale out from there.
What a real test costs
A credible test is not one machine. It is a coordinator plus several generators, ideally in more than one country so you can see how much of your latency is geography.
Say four 16 GB generators and one 4 GB coordinator, running for four hours, with two generators in Frankfurt and two in Bucharest:
| Role | Plan | Location | Hours | Cost |
|---|---|---|---|---|
| Generator ×2 | VPS IV KVM (16 GB) | Frankfurt | 4 each | €0.29 |
| Generator ×2 | VPS IV KVM (16 GB) | Bucharest | 4 each | €0.18 |
| Coordinator | VPS II KVM (4 GB) | Frankfurt | 4 | €0.05 |
| Total | €0.52 | |||
Fifty-two cents, before VAT, for a five-machine distributed load test. Even if you are generous with the setup time and leave everything running for a full working day, you are still under two euro. There is no version of this where the infrastructure cost is the reason you did not test.
A workable procedure
Raise the limits before you start
A fresh Linux image will not let one process hold fifty thousand sockets. Before anything else, raise the open-file limit for the user running the generator, widen the ephemeral port range, and shorten socket reuse timeouts. If your first run plateaus at a suspiciously round number, this is almost always why.
Test the generator against itself first
Point the load tool at a static file served from localhost on the generator and ramp it. Whatever rate that produces is your ceiling for that machine. Anything you measure against the real target above that number is measuring the generator.
Ramp, do not slam
Start at a fraction of the target, hold, step up, hold. A step ramp tells you where behaviour changes; an instant thundering herd tells you only that something broke. Watch the target's own metrics at the same time — a load test with no server-side observability is a stopwatch, not a diagnosis.
Generate from more than one country
This is the part people skip, and it is the part that produces the useful finding. Run the same scenario from two of our locations and compare the distributions. Our five VPS countries are Romania, Poland, the Netherlands, Germany and Austria, so you can put generators in Bucharest and Amsterdam and see immediately how much of the tail is distance and how much is your application. The per-location prices are all in the location pricing post, and the network page describes the connectivity.
Delete everything when you are done
Say it plainly: powering a VPS off does not stop the bill. A stopped virtual machine still holds its RAM, its disk and its IP on a node, and it bills as normal. Destroying it is what stops the meter. The teardown step belongs in the same script as the setup step, or you will find four idle generators next quarter.
Testing the thing you are about to launch
Load testing pairs naturally with a launch. If the event you are preparing for is a flash sale or a campaign spike, our post on VPS sizing for flash-sale traffic covers the other half — what to do with the answer once the test has given you one. And if the test is telling you the target is simply out of headroom, the same six plans scale up to 32 GB and 8 vCPU, or a dedicated server takes over where a VPS stops making sense.
The billing details, once
- Hourly services draw down prepaid account credit, once per clock hour, at the top of the hour in UTC.
- The hourly rate is the plan's monthly price divided by 720 hours. Thirty days of uptime costs the monthly figure; four hours costs four hours.
- No setup fees. Prices are euro, before VAT where it applies.
- If credit runs out, there is one hour of grace before suspension and 48 hours before termination, with warnings first and optional automatic top-up.
- A monthly summary invoice closes the period, so your accounting gets a document rather than a ledger dump.
Run one this week
Pick two 16 GB machines on the hourly VPS page, one in Frankfurt and one in Bucharest, and point them at your staging environment for an hour. Full root on a real KVM virtual machine, so nothing in the kernel or the network stack is off-limits to you. Total spend, about six cents.