I tested the performance of the NAS using Intel NASPT 1.7.0, which replays prerecorded traces of different tasks. I installed Western Digital Caviar Black WD5001AALS hard disk drives (500GB, 7200rpm, 32MB buffer, SATA II) and also tried Samsung HD753LJ drives provided by the NAS manufacturer.
The client was a PC with an Intel Core 2 Duo 1.8GHz processor, 4GB of system memory, a Gigabit Ethernet controller on PCIe 1x bus, and 32-bit Microsoft Windows Vista. The NAS was connected to the LAN via a Gigabit Ethernet router. Jumbo Frames technology was enabled on every device that took part in the testing (but certain tests were performed with Jumbo Frames off as indicated in the description of each test). I built a RAID array and created a shared folder on it and used that folder for the tests. The NAS’s other parameters were left at their defaults, so most of the additional services were turned off. These services may have a big effect on the NAS’s real performance (for example, when you are seeding 20 torrents on a fast Internet-connection).
In the first test I compare the performance of the NAS with HDDs from two brands working in five-disk RAID0 and RAID5 arrays. These array types provide the largest storage space. I use firmware version 844.
As you can see, the difference is no bigger than a couple percent. So, there is no point in choosing highest-performance HDDs for such NASes. You should instead choose HDDs basing on such factors as storage capacity, noisiness, heat dissipation and power consumption.
In the second test the NAS is tested with different firmware versions and with Jumbo Frames turned on and off. DSM 2.1 844 and DSM 2.2 914 are used.
The firmware update affects one subtest only: FileCopyToNAS. Jumbo Frames technology has a positive effect on the NAS’s performance in the streamed writing templates.
The third test helps us see the difference between the disk controllers employed in the NAS. To remind you, the fifth HDD and the eSATA port are connected via a SiI3132 whereas the first four HDDs are connected via a 88SX7042 chip. So, I test two single-disk arrays in the first and fifth bays. The other parameters in this and subsequent tests are: firmware 914, Jumbo Frames on, WD disks.
This test shows that the speed of one disk almost does not depend on the disk controller. Marvell’s controller is a mere 2% ahead of Silicon Image’s, despite the more progressive bus.
And now I will test different RAID arrays. The fourth diagram shows the performance of RAID0 which is fast but not fault-tolerant.
When the array grows up from one to two disks, we can see a considerable performance boost. But there are almost no performance benefits when we add even more disks to the array. The maximum read and write speeds are 79MBps and 84MBps in this mode. The DS509+ is somewhat better than the Atom-based NAS we have tested earlier: the QNAP TS-439 had maximum read and write speeds of 76 and 73MBps, respectively.
Well, it is fault-tolerant array types that are more appropriate for this type of NASes. Let’s take a look at the next diagram.
It is clear that the CPU load affects the performance of the RAID5 array at writing. The write speed goes down to 63MBps at high CPU load. In the RAID6 mode the write speed is only 53MBps. The read speed doesn’t change much, though. It still equals the maximum of 78MBps. The Atom N270 working at a considerably higher frequency is slower with its read and write speeds of 74 and 54MBps, respectively.
Surprisingly, the Freescale mpc8533 processor with PPC architecture and 1.06GHz clock rate proves to be faster than the good old x86 at 1.6GHz, even considering the differences in software. Synology’s NAS is very good from this aspect.