Testing the NanoPi NEO and the NanoPi 2 Fire

Up and Running

CPU performance for current small computers is no longer a topic of discussion. Four high-clocked cores offer more computing power than is generally required. Therefore, I will omit the corresponding benchmarks here. Basically, it is almost impossible to access all of the available power. Even the RPi3 cannot perform extensive compiling tasks because of inadequate storage. However, it is often more important for server applications to simply have the cores ready for processing requests.

The I/O performance is far more important as a criteria for headless systems than CPU performance. Because data rarely originates in main memory, I/O performance is especially relevant to interaction between the network and hard drive, which connects to both of the NanoPis through a USB 2.0 port. Because the NanoPi NEO only has a 100Mbps Ethernet connection, the throughput is in the range of 10 to 12MBps at best. This is also the case for a Rasp Pi. Ethernet constrains the throughput on both computers. On the other hand, the NanoPi 2 Fire, with its Gigabit Ethernet, has a theoretical limit of around 35MBps over USB.

Various data intensive applications such as socat, scp, and rsync served as candidates in the test scenario. I omitted a test using iperf. As is all too often the case, the theoretical claims of transfer performance do not conform to reality. SCP and Rsync, on the other hand, require a lot from the CPUs because of the encryption that occurs during transfer and reveal whether dips in transfer performance occur, as with the Banana Pis. The tests also show whether the USB jacks provide sufficient power for a level of consumption that goes beyond the specification.

Performance

Table 2 summarizes the performance results, including numbers for a current RPi3 and Banana Pi BPI-R1 for purposes of comparison. A SATA SSD was mounted on the existing jack of the Banana Pi, whereas the other two computers used an M.2 SSD in the USB 3.0 housing for bulk storage. The Pi Zero was intentionally not included in the testing because it lacks the necessary interfaces. I could have retrofitted these had I chosen to do so. However, the computers I did select for my server scenario test play in a much different league. The Pi Zero simply can't measure up.

### Server (Test Device)
$ socat TCP-LISTEN:2222 /<Path/to/file>
### Sender
$ time socat TCP:<target host>:2222 /<Path/to/file>

Under ideal circumstances, the write performance over USB 2.0, which I measured with dd by circumventing the cache, uses all of the NEO USB interface. This is a very good result. The NanoPi 2 Fire turns out to be significantly weaker in this regard. As predicted, the SATA interface on the Banana Pi delivers the best results.

The picture looks different, depending on the particular case, when data is sent over the Internet. Generally, it would be fair to say that the network interface on both the NanoPi NEO and the RPi3 constrict the data rate. The NanoPi 2 Fire offers much better performance in this regard, but it still lags far behind a throughput rate that can genuinely be measured in gigabits. However, the rsync example also shows that it is necessary to look at the actual results for each case.

As expected, the power supply for the 2.5-inch hard disks is usually sufficient. The only thing that overwhelmed the NanoPis was an old Transcend disk. The root of this problem lies with the disk, which also won't start up on a Rasp Pi.

Power consumption for the NanoPi systems was not anything out of the ordinary. Unlike Rasp Pis, the NanoPis do use less power once they are shut down. This is a long-time and well-known Rasp Pi weakness. Even though the Pi Zero was explicitly constructed as a low-power system, you can use a few tricks to persuade a NanoPi to reduce energy consumption to the same level by switching off the GPU and CPUs and reducing the clock frequency for the system. Because this occurs in userspace, it is possible to adapt the NEO to the load dynamically. The Armbian community is working steadily on this topic and offers relevant information and instructions on their website.