Last week, I looked at whether partitioning or formatting USB flash drives in other ways made a difference, and I found out that the stock FAT format tends to produce the best results overall. Today I’m testing striped RAID arrays using the same flash drives in 2, 3, and 4 drive configurations.
Some weeks back, a local outlet store had 8 GB USB flash drives for just $1.50 each. I grabbed four. When I went back a few days later, none were left.
The flash drives look like the minions from Despicable Me 2, but for $1.50, they were a steal. The flash drive itself appears to be sourced from PNY, a brand you’ll often find in retail stores.
Other than capacity and the fact that it’s a USB 2.0 device, I haven’t been able to find any more technical specifications.
Round 1: PowerPC – Power Mac G5
For the PowerPC set of tests, I used my 2.3 GHz Power Mac G5 Dual and ran Xbench 1.3 under OS X 10.5 Leopard. The first set of results is for a single USB flash drive. The second is for a striped RAID array using three of these USB 2.0 flash drives set to use 16 KB segments.* Two are plugged directly into the Power Mac G5, and the third is plugged into a port on the monitor, which has a 4-port hub. (Without disconnecting the keyboard, also plugged into a monitor port, there was no accessible spot available to plug in a 4th flash drive because there just isn’t enough room above this vertically oriented display.)
* I tried other settings, but they produced worse results than 16 KB.
The partition type is FAT and the format is Mac OS X Extended (Journaled). Here are the results (higher numbers are better):
Format: FAT RAID16
Disk Test 0.31 1.06
Sequential 10.18 13.86
Uncached Write, 4K blocks 7.05 13.87
Uncached Write, 256K blocks 6.68 10.75
Uncached Read, 4K blocks 14.86 10.40
Uncached Read, 256K blocks 29.22 36.56 18.38 MBps
Random 0.16 0.55
Uncached Write, 4K blocks 0.04 0.15
Uncached Write, 256K blocks 0.84 1.25
Uncached Read, 4K blocks 466.43 431.04
Uncached Read, 256K blocks 80.35 96.82 17.96 MBps
Overall, this RAID setup is over 3x as fast, easily winning six out of eight benchmark tests. Maximum read speed is 18.38 MB per second with 16 KB segments. That translates to 147 Mb per second, which is still far below the 480 Mb per second bandwidth of USB 2.0 – but over 20% better than the 122 Mb per second of the single-drive test.
The way USB works, no device can use more than about 70% of total bandwidth, which comes to 336 Mbps. At least we’ve hit 44% of that limit this time, a big step up from 36%. That said, the implementation of USB on PowerPC Macs just doesn’t hold up compared with Intel-based hardware.
Round 2: Intel – Core 2 Duo Mac mini
The next set of tests were run on my 2.0 GHz 2007 Mac mini with OS X 10.6.8 Snow Leopard. The flash drive was plugged into a USB 2.0 port in my monitor, since accessing the rear of the Mac mini is very difficult with my current setup.
This time I was able to use four USB flash drives, all plugged into the 4-port hub in my monitor, and create a striped RAID array. This is compared with the original single-drive FAT results. As with the tests on the Power Mac G5, each flash drive was partitioned as FAT and formatted as Mac OS X Extended (Journaled) and RAID set to 16 KB segments.
Because these are all connected via hub to the same USB 2.0 port, performance may not be as high as it would be were the flash drives plugged into separate ports on the Mac mini itself.
Format: FAT RAID
Disk Test 0.32 1.82
Sequential 12.78 22.64
Uncached Write, 4K blocks 8.64 23.09
Uncached Write, 25K blocks 8.11 16.10
Uncached Read, 4K blocks 18.04 16.87
Uncached Read, 256K blocks 53.94 83.07 41.75 MBps
Random 0.16 0.95
Uncached Write, 4K blocks 0.04 0.27
Uncached Write, 256K blocks 0.83 1.99
Uncached Read, 4K blocks 698.27 623.15
Uncached Read, 256K blocks 135.30 213.05 39.53 MBps
This time, maximum throughput was 41.75 MB per second (336 Mb per second). That’s the maximum theoretical throughput for a single USB 2.0 port., and that’s a 52% improvement over the best single flash drive throughput of 27.40 MB per second (219.2 Mb per second).
Overall, the disk benchmark score is nearly 6x that of the single-drive test! Intel-based Macs have a real edge in USB 2.0 performance vs. PowerPC, but even then these flash drives don’t achieve half the bandwidth of USB 2.0.
Round 3: Intel – Core 2 Duo MacBook
The third set of tests were run on my 2.0 GHz Late 2008 Aluminum MacBook with OS X 10.9.5 Mavericks. For the RAID test, two flash drives were plugged into the MacBook’s only two USB 2.0 ports, making this a very practical test for this hardware. System Information shows these two ports use separate controllers. Because these are plugged directly into the MacBook and into separate USB controllers, I would expect the best performance among the three striped RAID tests I’ve run.
As above, the flash drives were partitioned as FAT and formatted as Mac OS X Extended (Journaled), and RAID was set to 16 KB segments.
Format: FAT RAID
Disk Test 0.32 0.77
Sequential 12.78 12.48
Uncached Write, 4K blocks 8.85 15.01
Uncached Write, 25K blocks 7.77 14.37
Uncached Read, 4K blocks 18.47 17.50
Uncached Read, 256K blocks 52.45 105.90 53.23 MBps
Random 0.20 0.39
Uncached Write, 4K blocks 0.05 0.11
Uncached Write, 256K blocks 0.82 0.99
Uncached Read, 4K blocks 700.10 670.99
Uncached Read, 256K blocks 130.68 244.04 45.28 MBps
Overall, the two-drive RAID array has a disk test score nearly 2.5x that of the single flash drive, and some individual tests show the striped RAID array more than doubling the score of one drive.
Maximum read speed of this two-drive array is 53.23 MB per second (425.8 Mb per second), which is 89% of the throughput possible using a single USB 2.0 port. By using two separate controllers, we get our best results yet – and with just two flash drives. Even better results may have been obtainable using two USB 2.0 hubs, one connected to each port, but that’s not really practical to lug around with a notebook computer. That said, it could achieve read speeds on the order of 80 MB per second.
The maximum read speed with our two-drive striped RAID array is 97% higher than with a single flash drive (which achieved 26.97 MB per second on this MacBook), coming very close to doubling performance.
Remaining Questions
What kind of performance would we see with a 3-4 drive stripped RAID array plugged directly into the Mac mini, a MacBook Pro, an iMac, or a Mac Pro? What kind of performance would we see if I had a Mac equipped with USB 3.0?
What is it about the uncached read tests with 4K blocks that makes RAID slower than a single drive?
Conclusion
This may sound kind of crazy, but creating a striped RAID array using USB 2.0 flash drives can achieve some very decent results at minimal cost – not that I expect you’ll find 8 GB flash drives for $1.50.
To the best of my knowledge, you can’t boot from a USB flash drive RAID array, but if you’ve got one of those new-fangled Macs where Apple doesn’t make it easy to replace the SSD, this could be an inexpensive way to add a decent amount of portable storage space. This would be especially true for MacBook Air models with two USB ports.
It’s not going to come close to the speed of SATA, but it’s could be a lot more affordable than most third-party SSD upgrade drives while adding minimal bulk – if you find drives that barely stick out beyond the USB port.
Keywords: #flashdrivespeed #flashdriveperformance #flashdriveformat #flashdrivebenchmark #usbthroughput #stripedraid #raidarray #flashdriveraidarray #flashdriveraid
Short link: http://goo.gl/zxDrak
searchword: flashdriveraid
