VPS - Benchmark Test

Benchmark tests certainly don’t tell you everything about the performance of a server. Especially when evaluating a provider, completely different criteria play a decisive role. Nevertheless, I wanted to benchmark a few smaller VPSs to see how big the differences are. Of course, the values only say something about the tested performance over a short period of time. The results give no indication of service, uptime or performance over a more extended period of time.

Which servers do I have that I could test right now?

Of course, this does not compare similar servers, but the result may still be enjoyable. All servers in this list cost less than €10 per month. I have also included my local Intel NUC BOXNUC7CJYH2 Celeron J4005(Amazon Affiliate Link) with G.Skill 2x4GB RAM (Amazon Affiliate Link) and a Crucial MX500 500GB SSD(Amazon Affiliate Link) in the comparison.

Test Preparation

I install a new Ubuntu 20.04 on the VPS (also on the Intel Nuc) and then connect to the server. I then update the server.

apt update
apt upgrade

Then I install the benchmark tools with which I want to carry out the tests.

Installation Sysbench

apt install sysbench

Installation 7zip

apt install p7zip-full

Install Geekbench

wget https://cdn.geekbench.com/Geekbench-5.4.1-Linux.tar.gz
tar xf Geekbench-5.4.1-Linux.tar.gz

Installation Phoronix Suite

apt install gdebi-core
wget http://phoronix-test-suite.com/releases/repo/pts.debian/files/phoronix-test-suite_10.4.0_all.deb
gdebi phoronix-test-suite_10.4.0_all.deb
phoronix-test-suite install john-the-ripper
phoronix-test-suite install nginx

Restart

At the end I restart the server once.

reboot

Test execution

Sysbench

I used Sysbench to test the CPU, RAM and SSD. Each test ran three times to reduce the impact of deviations. I ran four different tests, with one, two and four threads respectively, regardless of whether multiple CPU cores were even present, for example.

sysbench cpu --threads=1 run
sysbench cpu --threads=2 run
sysbench cpu --threads=4 run

Here I have noted the events per second

sysbench memory --threads=1 run
sysbench memory --threads=2 run
sysbench memory --threads=4 run

Here I have noted the operations per second

sysbench fileio prepare

This is not a test, I only ran it once, but I noted the write speed anyway

sysbench fileio --threads=1 --file-test-mode=seqwr run
sysbench fileio --threads=2 --file-test-mode=seqwr run
sysbench fileio --threads=4 --file-test-mode=seqwr run

Here I have noted the written, MiB/s

sysbench fileio --threads=1 --file-test-mode=rndrw run
sysbench fileio --threads=2 --file-test-mode=rndrw run
sysbench fileio --threads=4 --file-test-mode=rndrw run

Here I have noted the read, MiB/s and written, MiB/s

7zip

With 7zip I ran the standard benchmark test once and compared the average MIPS for compression and decompression. This test also ran with one, two and four cores.

7z b -mmt1
7z b -mmt2
7z b -mmt4

Geekbench

In the unzipped folder I executed the Geekbench script.

cd Geekbenchfolder
./geekbench5

Phoronix test suite

With Phoronix I executed both previously installed tests.

phoronix-test-suite run john-the-ripper
phoronix-test-suite run nginx

I selected Blowfish for John-the-ripper.

Test result

Now, we come to the exciting part: the evaluation.

Sysbench

CPU

In the CPU test with Sysbench and the limitation to one thread, my Intel Nuc achieved the best results. As soon as two threads were allowed to be used, it was narrowly overtaken by the Netcup root server. The 1blu server achieved a significantly better result with four threads than the rest.

Memory

The memory test had a different winner depending on the number of threads. With one thread, it was the small Hetzner VPS with the Intel CPU. With two threads, it was the Racknerd server, closely followed by the Intel Nuc. With four threads, the memory of the Contabo server was able to process the most operations.

Fileio

The 1blu server achieved the best results in terms of write and read speed. Unfortunately, I don’t have a second server with this provider to test whether this was an outlier. To make it easier to compare the remaining test results, I have created another diagram without 1blu.

Without 1blu for better comparison. Grouped by threads and test mode, diagram created with Microsoft Office Excel](@assets/images/blog/vps-benchmark-test/images/image-11.png)

Netcup’s root server came a distant second, followed by the Hetzner VPS with the Intel CPU. I am surprised at the large gap between the Hetzner VPS with the Intel CPU (CX11) and the Hetzner VPS with the AMD CPU (CPX11), as I would assume that both use the same type of hard disk.

Incidentally, I also tried this test on a Raspberry Pi 3B+(Amazon Affiliate Link) with one GB RAM and a 32 GB Samsung Evo Plus MicroSD(Amazon Affiliate Link). All results were less than 10 MiB/s, which is why I would not use a Raspberry Pi with a MicroSD card for processes that require many writes, the I/O is too slow.

7zip

-mmt1

In the 7zip benchmark with only one thread, the Hetzner CP11 server achieved the best results.

-mmt2

The 1blu server achieved the most MIPS from two threads.

-mmt4

With four threads, the 1blu server was able to extend its lead.

Geekbench

In the Geekbench test, I also included three smartphones with Android in the comparison.

The root server from Netcup achieved the highest single-core score. In the multi-core score, it was beaten by the 1blu server.

Phoronix test suite

John-the-ripper

The 1blu server also achieved the best result in the John-the-ripper test in the Phoronix test suite. The Racknerd server took second place here.

Nginx

In the Nginx server test, the servers with AMD CPUs achieved the best results. In 3rd place was the Hetzner CPX11, in 2nd place the Netcup VPS 1000 G9 and the Netcup root server 1000 G9 was able to process the most requests.

Netcup discount vouchers

The test results were not sponsored by anyone. New customers at Netcup can use my netcup-affiliate-vouchers to get a server for less. This gives you a discount on your order and I receive a commission.

In my opinion, in most cases it is worth the extra charge for a root server if you have no problem committing to a contract for 12 months.

Raw data

I have also copied the results of the benchmark tests into a Google spreadsheet, which can be viewed here.

Which provider did you have the best experience with? I would be happy to receive your comments.