I recently had the chance to evaluate a couple of iSCSI solutions for a pair of clients looking for a massive network based storage solution. Rather than attach large external hard drives to several workstations on the network, we wanted to consider a SAN solution. In this case, a single drive chassis with at least 4 RAIDed hard drives. I had previously evaluated the Drobo Elite from Data Robotics Inc. This time I took a look at the TS-509 Pro Turbo NAS from QNAP.
The TS-509 is a NAS/SAN solution that offers dual gigabit network ports that can be used for iSCSI connectivity. The device is completely self-contained. Simply supply it with power and attach it to the network and, once configured, it functions autonomously. The chassis supports up to 5 SATA 3.5” hard drives. For my testing, I installed 3 Hitachi Deskstar 2TB SATA drives stripped together in a level 5 RAID. The drives were 7200RPM with 32MB buffers.
My network centers on a 24 port gigabit D-Link switch. But since this is an unmanaged (consumer quality) switch, I wanted to make sure the fairly low-end switch would not hinder test results. When I tested the Drobo Elite, I attached each of the two test computers directly to each of the Drobo’s gigabit Ethernet ports. I wasn’t able to do this with the TS-509, so each of the network ports was connected directly to the switch via a pair of Cat-5e network cables.
The first test computer was a first generation quad core 2.66GHz Mac Pro. The other connection was to a 2.53GHz Core 2 Duo based Unibody MacBook Pro. Both have onboard gigabit networking. Both were running OS X 10.6.3. Both of their network settings were configured manually to 1000baseT, full duplex. Since the TS-509 does not support jumbo frames, the default MTU was left at 1500.
Each drive was tested locally in order to establish the baseline performance of each computers internal drives. These numbers also gave me something with which to compare the overall performance of the TS-509. Ideally, the tests over iSCSI would yield scores comparable to a computers internal drive mechanism.
MacBook Pro Internal Boot Drive:
Mac Pro internal 2TB Non-Boot Drive:
The remainder of the tests were performed from the MacBook Pro using SpeedTools Utilities Pro from Intech. Large transfers were moving either to or from the TS-509 via the Mac Pro. While the massive transfer was under way, I wanted to find out what another machine on the LAN could expect for performance. For the sake of testing, 1 set of files contained many large media files of about 10GB each. Another test was entirely very small web design html, gif, and jpg files (500,000+).
MacBook Pro Test Results: Large File Write
(MacBook Pro running bench test while massive large file transfer is being written to the TS-509 from the Mac Pro)
MacBook Pro Test Results: Large File Read
(MacBook Pro running bench test while massive large file transfer is being read from the TS-509 to the Mac Pro)
MacBook Pro Test Results: Small File Write
(MacBook Pro running bench test while massive small file transfer is being written to the TS-509 from the Mac Pro)
MacBook Pro Test Results: Small File Read
(MacBook Pro running bench test while massive small file transfer is being read from the TS-509 to the Mac Pro)
The speeds returned by SpeedTools showed less of a performance hit than I expected when a massive file transfer was already under way between the Mac Pro and the TS-509. Ideally I would have had a couple of additional computers moving data to and from the NAS while I ran the tests, but I didn’t have any other gigabit Mac’s handy and didn’t want to throw Windows into the mix and muddy the waters.
There is one more consideration when using an iSCSI solution. All of the data is moving across the network port of the computer. This puts a considerable demand on the processor. On the Core 2 Duo MacBook Pro, while doing a sustained file transfer, I was seeing about 25% constant usage on each of the two processor cores. On the quad core Mac Pro, I was seeing about 10% processor usage. This is directly inline with the processor usage I see when I do a large file transfer to a conventional network file server. USB 2.0 suffers a simpler processor related performance hit while FireWire transfers offload the overhead to the FireWire controller leaving the processor virtually unaffected.
Having already evaluated the Drobo Elite, I was impressed at how well the TS-509 stacked up against it. While the Drobo Elite was a powerful solution, it had few options in the way of customization and optimization. The network admin simply sets it up and lets it run. The TS-509, on the other hand, has many configuration options. Some of which could seriously increase performance. For example, there are several options for ganging the two gigabit network ports together for either redundancy, or optimizing throughput. I lacked a layer 2 switch to take advantage of more sophisticated networking options, but its nice to know that the hardware has such flexibility for those looking to achieve either redundancy or greater network optimization.
In the end, the performance of the TS-509 was comparable to the Drobo Elite. Granted, the TS-509 offers only 5 drive bays while the Drobo Elite has 8. But with the Drobo solution priced at 3 times that of the QNAS solution, my client decided to deploy the TS-509.