The Old Cloud – CPU and Disk Performance

In the previous, I have taken a first look at Scaleway’s bare metal servers that are around 10 years old at the time of writing. Compared to other companies offering more recent hardware, their prices are significantly lower for some of their configurations. Despite the dated hardware, the offer is still interesting to me, as my main two requirements is RAM and disk space. Performance is only a secondary requirement, as my services are only used by a few people concurrently. But still, it would be good to know how Scaleway’s old servers compare against other offers. The two most important data points for me are CPU and SSD performance. So here we go:

CPU Performance

To measure CPU performance, I’ve been using my ffmpeg encoding speed test. the Intel CPU E3-1230 V2 (ca. 2012) with 4 cores and 8 threads runs my test in 9 minutes and 15 seconds.

In comparison: My 2023 Lenovo T14 notebook with a 13th Gen Intel i5-1335U CPU runs the same test in 5 minutes and 12 seconds, while my bare metal server with an AMD Ryzen 5 3600 CPU (ca. 2021 architecture) runs the same test in 4 minutes and 5 seconds.

In other words. The old Scaleway server has around 50% of the performance of contemporary hardware. I had no idea what to expect before running the test, but from a performance point of view, that’s still good enough for me.

Disk Performance

Now lets come to the SSDs. Here, I have to admit, performance does count to some degree, as I often have to transfer large amounts of data, e.g. while compressing virtual machine images. To test real world performance, I created a 62GB file with random data (dd if=/dev/urandom of=rnd_data-62GB.img bs=1024 count=60751872). Surprisingly, creating the file on the old Scaleway hardware took only half the time compared to creating the same file on Hetzner’s more modern bare metal server. So it looks like the random generator on that AMD CPU is much slower than on a 10 year old Intel CPU. But creating files with random content is not a real world workload, so I’ll not dwell on this.

What I was after was how fast that 62 GB file can be copied on the system. On both bare metal servers, I use two disks and ZFS to create an encrypted file system across both physical disks. While both servers have SSDs, the Hetzner server has two 512 GB NVMe SSDs, while the Scaleway server has two 1 TB SATA SSDs that use the slow SATA 2.0 mode with a maximum 300 MB/s data transfer rate rather than SATA 3.0, which would have allowed 600 MB/s. It would have been surprising to see similar data transfer rates and indeed, the difference is quite significant:

On the Scaleway system, copying the 62 GB file with random data took 6 minutes and 23 seconds. On the Hetzner system, the same action took 2 minutes and 59 seconds. That’s over twice as long, and, I have to admit, that hurts a bit.

Summary

Overall, the 10 year old bare metal server in Scaleway’s data center has about 50% the CPU and disk performance of my bare metal server in Hetzner’s data center. But it comes at half the price per month and has twice the amount of SSD storage. An interesting balancing act.