Data Dance

Netgear

The Netgear model has four disk trays and also allows the configuration of larger volumes. Like the Thecus model, it has an LCD display, but unlike it, no pushbuttons. You can thus quickly discover what IP address the device uses, but you cannot set the IP. Otherwise, the hardware equipment is good; however, the device is not cheap.

The Linux operating software, dubbed ReadyNAS OS here, left the testers somewhat ambivalent. On the one hand, it was the only device in our lab to use Btrfs by default as its NAS filesystem, and it passes on the benefits – such as the ability to grab snapshots – in full to the user. Elsewhere, however, the device is still somewhat immature. It produces completely cryptic error messages (e.g., 1013050001 pre_proc_add_share() failed with status=1 ), which we hope convey something to the developers but are virtually useless to users.

When attempting to download the ReadyNAS Remote Client on Linux, we were incorrectly offered the version for Mac OS – apparently, because there is no Linux version. This also means that Linux users cannot access their NAS from the outside if it resides behind a NAT router.

Additionally, problems with the GUI sends you through endless loops; for example, if you click Cloud in the top menubar of the main menu, you end up on a page that offers you three services: ReadyCloud , ReadyNASReplicate , and ReadyDROP . If you choose ReadyCloud , you are guided through device detection (Discover ) and the Manage link, which are the only choices, then back to the main menu, where you can start over again. This doesn't make any sense, and although this annoyance does not keep you from using the device, but it does show some room for improvement.

Home-Grown

When looking at the test candidates some people might be asking: Can't I build something like this myself? Sure, you can. With a SoC like the Raspberry Pi, it could even have similar physical dimensions. The commercially available devices all take advantage of the free Linux operating system and rely on its standard tools; there is no wizardry involved here.

However, a hobbyist with a home-built device would have to do without a few features. You would not have a GUI for single-click installation and graphical configuration of selected applications, and the device could not be programmed without unreasonable overhead. Additionally, you would find it difficult to run services such as Dynamic DNS or the externally accessible internal server that allows users to connect from the web. An appealing case for the computer and hotplug-capable disks would be challenging, at least. Also, if you have problems, you would not be able to call the NAS manufacturer for support.

Thus, you probably gain little through DIY. Certainly some applications that you could imagine on a NAS have not yet found their way into the manufacturers' repositories, but these are special cases, and you might be able to solve them on a standard device after installing SSH access. On the other hand, doing so could void your warranty. For example, Netgear reserves the right to require a factory reset in case of support if the admin has enabled SSH access.

Benchmarks

For the main benchmarks, we turned to the Intel NAS Performance Toolkit (NASPT) [1], which was developed by the chip manufacturer for comparing NAS appliances, and which recreates workloads typical of applications in the home or small office. These include video playback with one, two, and four parallel streams (Figure 1), video recording, working with Office applications (Figure 2), video rendering, file transfer (Figure 3), or copying directory trees (Figure 4).

Figure 1: The differences are not so massive for video playback, which is mostly sequential reads. Only the Thecus device lags behind a bit with four parallel streams.

Figure 2: The two devices with Atom CPUs lose out with typical office jobs.

Figure 3: When copying files, the Thecus competitors take precedence.

Figure 4: During directory tree copies, the appliances have to write the files and create new subdirectories. The Atom group loses ground here.

The NASPT aims to measure performance from the end user's perspective, which – instead of just running some purely synthetic benchmarks – does justice to caching effects at the operating system level and CPU performance. (See the "How We Tested" box for more details.)

CPU performance test results especially proved a decisive factor: The Atom CPU in the Thecus achieved a CPU Mark [2] value of 667, which is just around two-thirds of the performance of the Celeron CPU (1047 CPU Mark) in the QNAP device. This difference is reflected in the application performance – not so much with sequential streaming of large files, but with small block sizes, several users working at the same time, or frequent metadata changes. For example, CPU performance considerably influences transfer rates while copying directories (Figure 4). In contrast, sequential reads are predominant in video streaming playback, and the differences here are relatively small.