Before running any benchmark
Run these commands only on a VPS you own or are authorized to test. They consume CPU time, disk I/O and network traffic. Prefer a quiet, non-production server, and check your provider's acceptable-use rules. Avoid heavy tests on a busy database or customer-facing server.
Record the OS version, vCPU allocation, RAM, test time and tool versions. Run each test three times under similar conditions. Compare results from identical commands instead of relying on one headline number.
Install the test utilities
sudo apt update
sudo apt install -y sysbench fio iperf3
sysbench --version
fio --version
iperf3 --versionThese package names apply to Ubuntu and Debian. Package installation may ask whether iperf3 should run automatically as a daemon; no permanent public test server is required for this guide.
Check CPU throughput with sysbench
Run the single-thread test first, then a test that uses all visible vCPUs. Both calculate prime numbers using the same workload. The output includes events per second, which is only comparable when the settings match.
sysbench cpu --cpu-max-prime=20000 --threads=1 --time=20 runsysbench cpu --cpu-max-prime=20000 --threads="$(nproc)" --time=20 runHow to read the result
Look for events per second and the thread count. Higher throughput for the same workload generally indicates better results in that synthetic test. Do not equate an extra vCPU with an extra dedicated CPU core; virtualization and CPU contention can change throughput.
Documentation: sysbench project.
Measure small-block reads with fio
fio can measure I/O operations per second, throughput and latency. This example creates a test file in your home directory, then runs a short 4 KiB random-read test with one synchronous I/O request at a time. It is intentionally modest and is not a maximum-IOPS stress benchmark.
mkdir -p "$HOME/vps-bench"
df -h "$HOME"
fio --name=prepare --filename="$HOME/vps-bench/fio-test.bin" --size=256M --rw=write --bs=1M --ioengine=sync --direct=1 --group_reportingfio --name=small-read --filename="$HOME/vps-bench/fio-test.bin" --size=256M --rw=randread --bs=4k --ioengine=sync --iodepth=1 --direct=1 --runtime=20 --time_based --group_reportingrm -f "$HOME/vps-bench/fio-test.bin"
rmdir "$HOME/vps-bench"How to interpret fio
IOPS measures operations per second; BWreports data throughput; and latency describes how long operations take. Lower latency is often desirable for databases and interactive applications. This particular command runs at queue depth one, so do not compare it to advertising that uses much higher queue depths.
RAID1 offers disk redundancy, but it is not a backup. Always keep separate copies of important data.
Documentation: fio reference manual.
Test throughput and routes with iperf3
A network benchmark needs two different machines. One VPS runs the iperf3 server and another connects to it. Testing against 127.0.0.1 measures local memory/network-stack performance, not your external uplink.
On a second server that you control, temporarily allow TCP port 5201 only from your testing IP address, then run:
iperf3 -sOn the VPS you want to test, replace the placeholder with that server's actual IP:
iperf3 -c YOUR_TEST_SERVER_IP -t 10
iperf3 -c YOUR_TEST_SERVER_IP -t 10 -RThe first command sends traffic toward the test server; -Rreverses the direction. When finished, stop iperf3 on the server with Ctrl+C and remove the temporary firewall rule. Your provider's traffic allowance may count transferred data.
Why latency is a separate measurement
Throughput alone does not tell you how responsive an application will feel. Test ping and routes from your chosen location to important destinations and check for latency variation or sustained packet loss. Some networks rate-limit or block ICMP responses, so a single missing ping reply does not by itself prove an application problem.
Documentation: ESnet iperf3 manual.
What to record for a repeatable VPS comparison
| Metric | Record | What can mislead you |
|---|---|---|
| CPU | Single/multi-thread events per second | Different thread counts or prime limits |
| Disk | 4 KiB read IOPS, average latency | Different block sizes, cache, queue depths |
| Network | Upload and reverse throughput | Remote-server limits and routing |
| Environment | vCPUs, RAM, OS, date and versions | Host contention, different test conditions |
Run repeated tests, record the median when practical, and evaluate the application itself. A web application, build job and database may respond differently to the same synthetic benchmark numbers.
Testing a Frankfurt VPS before you order
ClusterHosting provides a Frankfurt network test pageand Looking Glass for ping, traceroute and MTR. Those tools can help you inspect routes from one Frankfurt test server before buying. They cannot guarantee identical results from every VPS family or datacenter.
Once deployed, use the commands above on your own instance to build a reproducible baseline. Keep the results with your application monitoring so you can diagnose future performance changes.