Performance - New Platform
To quantify how much the current NAS Chart test machine was limiting the Pro's performance results, I ran a series of tests on the newer machine that I have been using in the Fast NAS series.
This is a Dell Optiplex 755 with an Intel Core 2 Duo E4400, onboard Intel 82566DM-2 PCIe gigabit NIC and 512 MB of RAM. I purposely kept the RAM low to see non-cached performance kick in before I hit the maximum 1 GB test file size. But given Pro's 1 GB RAM, I ran the tests out to a 4 GB file size.
Figure 10 shows three test runs of RAID 0 write performance with a gigabit LAN connection:
- Darth, XP SP2, Perf Tweaks - New testbed running Windows XP, SP2. ReadyNAS Pro System performance options set to disable all journaling, disk cache enabled and fast CIFS writes enabled.
- Putti, XP SP2, Perf Tweaks - Same as #1, but running on current NAS testbed machine
- Darth, Vista SP1, Perf Tweaks -Same as #1, but testbed machine runing Vista SP1
The 66.9 MB/s 1000 Mbps PCI line is a reference line for maximum LAN throughput measured between two PCI-based gigabit NICs. I didn't plot a reference line for the same measurement between two PCIe NICs (113 MB/s) because it would have compressed the plot vertical scale.
Figure 10: Testbed comparison - RAID 0 write - 1000 Mbps LAN
As expected, Figure 10 shows that the faster testbed machine achieves higher throughput from the Pro. Averages across the tested filesizes are 68.3 MB/s, 69.9 MB/s and 47.4 MB/s for test runs 1, 2 and 3 respectively. So I got an almost 50% average throughput gain by using the faster iozone test machine.
The best average throughput was produced using Vista SP1, but mostly by the boost at smaller filesizes. You can see that XP SP2 actually beat out Vista SP1 for file sizes 128 MB on up.
RAID 0 read results are shown in Figure 11, which show average speeds of 61.1 MB/s, 52.4 MB/s and 44.0 MB/s respectively for the three runs. The best case gain in average throughput this time is just shy of 40%, with Vista leading the way across all tested file sizes.
Figure 11: Testbed comparison - RAID 0 read - 1000 Mbps LAN
Of course, the question in many of your minds is how the Pro compares to the QNAP TS-509 Pro, which has been the subject of a long series of performance experiments by Dennis Wood and others over in the Forums.
While I still had the TS-509 Pro here (believe it or not, vendors don't leave all these expensive beasties here for me to play with indefinitely), I somehow had the foresight to run a few tests using the "Darth" testbed. That data is presented in Figures 12 and 13.
All tests were run on the same faster testbed machine, using Vista SP1. This time, however, the plots show the Pro configured in XRAID-2 with three drives, then five drives. The TS-509 Pro had all five drives configured in a single RAID 5 array. Jumbo frames were not used because the Vista drivers don't support them for the Intel onboard NIC. The QNAP data ends at 1 GB because I didn't have that much foresight!
Figure 12: QNAP TS-509 Pro comparison - RAID 5 write - 1000 Mbps LAN
Write performance is very similar for the Pro and TS-509, with even a similar throughput fall-off as cache effects die out. Although the results show the TS-509 slightly beating out the Pro, I think you would be hard-pressed to notice a difference in real-life use. It's also good to note that adding more drives to the Pro in XRAID-2 doesn't significantly impact write performance.
Read performance in Figure 13 is a bit more interesting. Both the QNAP and Pro start out with essentially the same throughput in the 65 to 70 MB/s range. But the Pro with 3 drives loses steam after the 64 MB file size, while with 5 drives holds up until after 256 MB. The QNAP maintains its steady throughput until after 512 MB, then really crashes to earth at 1 GB.
Figure 13: QNAP TS-509 Pro comparison - RAID 5 read - 1000 Mbps LAN
At NETGEAR's request I ran a quick IOMeter test on the Pro in XRAID-2 with 5 drives, but only managed to hit 95 MB/s for writes and 71 MB/s for read, far short of the 100+ MB/s that NETGEAR measured using a Intel Quad Core 2.66 GHz test machine with 2 GB of RAM, running Vista SP1.
Updated 11/21/2008: Explanatory note added
As a final test, I ran write and read drag-and-drop tests using the same "Darth" machine used in the Figure 12 and 13 tests. Of course, a machine-to-machine filecopy brings the performance of test machine's disk subsystem into play. For the "Darth" machine, the drive is a Seagate Barracuda 7200.10 80GB SATA (ST380815AS).
Certainly not the fastest SATA drive out there and with only 8 MB of cache, I'm probably testing the "Darth" machine's performance more than the Pro. In fact, NETGEAR says that to fully utilize the Pro's performance, you need to use a "large ramdisk or a RAID 0 with 3 striped disks". So transferring a 4464 MB folder containing a ripped, uncompressed DVD yielded 50.7 MB/s write and 48.5 MB/s read.
I pulled the Vista Drag-and-Drop filecopy, Iozone X-RAID2 (Figures 12 and 13) and IOMeter results into Table 1. My take-away from this is that it appears that iozone's results, which are not limited by the hard drive speed of the machine it runs on, are more indicative of real-world file copy performance (at least as defined by Vista SP1) than IOMeter's results.
|Test||Iozone X-RAID2, 5 drives
(avg 32 MB - 4 GB)
|Vista SP1 DnD||IOMeter|
Table 1: ReadyNAS Pro X-RAID2 performance comparison
Knock yourself out running your own comparisons using the NAS Charts.
There is little doubt that NETGEAR has really stepped up its game with the Pro and its switch from the aging Infrant processor platform to Intel-land. While the Infrant processor got the ReadyNAS line to where it is today (part of NETGEAR), it was long past time for NETGEAR to move off that proprietary dead-end and onto a platform that offers a much wider array of price / performance options.
The Pro deserves its status as one of the top-ranking products in our NAS Charts. But I'll leave it to the fanboys to argue whether it or the QNAP TS-509 Pro (or Intel SS4200-E or Sans Digital MN4L+ or Synology DS508) is really the top dog. From what I can see, you'll probably see more difference in performance among the group due to client OS, memory and networking interface than from the NASes themselves.
What's not subject to debate is that the Pro is the most expensive 5+ drive NAS available...at least for now. NETGEAR plans to release a "Pioneer" diskless model ($1950 MSRP, ~$1200 street) once the current tight availability situation improves (December earliest, more likely early 2009).
The Pioneer version will have most of the features of the diskful models except snapshotting, Active Directory, and VLAN. (A full PDF model comparison list is here.) But even at $1200, the "Pioneer" Pro will still be the most expensive BYOD NAS; even more than the upcoming seven-drive Thecus N7700 ($1099 MSRP).
The ReadyNAS line has always had a solid feature set and user-friendly automatic RAID. And its move under the NETGEAR umbrella only enhanced its value proposition. What was lacking was performance, and now that has been addressed.
In the end, the Pro is going to give NETGEAR a shot at getting business back from the lesser-known and smaller (and not as widely available) "QST" (QNAP, Synology, Thecus) challengers that have been stealing performance-focused buyers away. Let the price wars begin!