Benchmark: our 4.54 EUR Lisbon VPS on a busy production node
Hosting benchmarks are usually run on an empty server, the day it is installed. We did the opposite: we ordered our smallest plan on a production node carrying normal customer load, and measured it.
The test setup
- Plan: the entry VPS, 2 vCores, 4 GB RAM, 100 GB SSD, 4.54 EUR per month.
- OS: Debian 13 from our standard template.
- Location: ONI datacenter, Lisbon, Portugal.
- Host: a production node with its usual customers on it, picked because its load is typical, not because it is quiet.
- Disk: local ZFS on SSD. Lisbon VPS can also run on our Ceph HA storage; this test used local ZFS.
- Date: 7 November 2026.
Deployment time
| Step | Time after the order |
|---|---|
| VM running, IP address assigned | 37 seconds |
| SSH login works, first boot setup done | 2 minutes |
CPU and memory (sysbench, 20 s per run)
The vCPUs are Intel Xeon Gold 6142 cores, passed through as cpu: host.
| Test | Result |
|---|---|
| CPU, 1 thread | 1,037 events/s |
| CPU, 2 threads | 2,063 events/s |
| Memory, 2 threads | 4,584 MiB/s |
Two threads score 1.99x one thread. On a host where vCPUs fight for the same physical cores, the second thread adds much less. Our CPU is not capped, and each host runs at most 3 vCPUs per physical thread.
Disk (fio, libaio, direct I/O, 4 GB test file, 20 s per run)
| Test | IOPS | Throughput | p99 latency |
|---|---|---|---|
| 4K random read, queue depth 64 | 108,000 | 423 MiB/s | 1.0 ms |
| 4K random write, queue depth 64 | 64,000 | 250 MiB/s | 3.1 ms |
| 1M sequential write, queue depth 16 | - | 424 MiB/s | - |
We left out the 1M sequential read result on purpose. It was 5.8 GB/s, but that came from the host's ZFS memory cache, not the disks, so it says nothing about storage. Random 4K reads also get help from that cache, which is what a database sees for its hot data in real use.
Latency from Lisbon (average ping, IPv4)
| Destination | ms |
|---|---|
| Madrid | 14.7 |
| Paris | 27.5 |
| Frankfurt | 33.5 |
| London | 34.0 |
| Amsterdam | 34.5 |
| Falkenstein | 39.0 |
| Stockholm | 52.5 |
| Warsaw | 53.6 |
You can run the same test yourself, without ordering anything, from our Looking Glass in Lisbon.
Downloads (curl, one connection)
| Source | Speed |
|---|---|
| Madrid | 70 MB/s |
| Roubaix (OVH) | 60 to 73 MB/s |
A single TCP connection over a long distance is limited by the far server and by latency, so these numbers depend on the other side as much as on us. Each VPS gets up to 0.5 Gbit/s and 20 TB of traffic per month, with no overage fees.
How to repeat it
These are the parameters we used. Run them on any VPS and compare.
sysbench cpu --threads=1 --time=20 run
sysbench cpu --threads=2 --time=20 run
sysbench memory --threads=2 --time=20 run
fio --name=rr --rw=randread --bs=4k --iodepth=64 --ioengine=libaio --direct=1 --size=4G --runtime=20 --time_based
fio --name=rw --rw=randwrite --bs=4k --iodepth=64 --ioengine=libaio --direct=1 --size=4G --runtime=20 --time_based
fio --name=sw --rw=write --bs=1M --iodepth=16 --ioengine=libaio --direct=1 --size=4G --runtime=20 --time_based
The plan
The tested plan is the entry VPS in Portugal: 2 vCores, 4 GB RAM, 100 GB SSD, 20 TB traffic, 4.54 EUR per month, billed by the hour. IPv6 is included, and a dedicated IPv4 address is 0.70 EUR per month.