SSD Benchmarks

Due to the inclusion of M.2 NVMe as a storage option, we decided to run the full set of benchmarks using a Micron 2450 NVMe SSD based around the Phison E19T DRAM-less controller to see if there was any benefit of swapping the mechanical drives for an SSD, especially at 10 Gbps. This is the same SSD that was using for the SSD caching benchmarks, and it should be pretty straightforward to compare the numbers.

CrystalDiskMark

We were quite surprised to see that the RAID 1 benchmarks are the ones that came out on top in CrystalDiskMark, but we're even more surprised that RAID 1 also beat the SSD. The SSD used is hardly a high-end model, but it was beaten in all the read tests. However, the SSD excels in the sequential write tests and shows that the mechanical drives are not ideal in scenarios when a lot of data needs to be written.

NAS Performance Tester

Here we're seeing a decent performance gain over the RAID 0 write performance, yet we're only at just over half the theoretical limit of 10 Gbps Ethernet. However, the read test shows the benefit of having an SSD in a NAS, even though it won't apply to all scenarios. At almost 865 MB/s we'd call that a pretty good result. We're not sure what was going on with the 2.5 Gbps write tests, but we ran it multiple times and ended up with similarly disappointing results. On the other hand, the read test is more or less maxing out the 2.5 Gbps interface.

Helios LanTest

Here once again the RAID 1 read tests outclasses everything else, but in the write test, the single SSD is nearly as fast as the RAID 0 array. The SSD should be much faster here, but it appears that there's some kind of bottleneck that prevents it from delivering optimal performance. All the response times are also similar to that of the RAID 0 setup. Using the slower 2.5 Gbps interface shows a small advantage to the SSD.

Game Data

Here the SSD excels over the hard drives except compared to the RAID 0 setup in the write test over 10 Gbps. However, it doesn't beat the RAID 5 setup with the SSD as a cache, and we can't really work out why, as the SSD ought to be just as fast, if not faster here. There appears to be some kind of bottleneck when just an SSD is used in some scenarios. Switching to the 2.5 Gbps interface, the SSD easily outperforms the mechanical drives, no matter the RAID level, but not by a huge margin by any means.

Single Large File

This is where we see some real benefits from using an SSD and over 10 Gbps it's the clear winner, even compared to the tests performed with the SSD acting as a cache for the hard drives. However, there appears to once again be some kind of oddity going with the 2.5 Gbps interface, as the write speed is slower compared to the mechanical drives here, yet the SSD is nearly hitting the limit of the 2.5 Gbps Ethernet interface in the read test.

Photos

With multiple smaller files, the SSD is still the champ, performing similar to the cached reads and outright beating the mechanical drives in the write test, as one would expect. Over 2.5 Gbps the write performance is on par with the RAID 0 setup, but it easily beats the hard drives in the read test.

We're seeing quite an odd behavior in some tests from the SSD, but we re-ran the tests to verify that the performance was indeed correct and ended up with very similar results each time. As such, we can only chalk down the performance peculiarities to an immature platform that needs some work.