An Array of Tiny Drives

By Loyd Case  |  Print this article Print

Six hard drives in a RAID array: It sounds like it would be noisy, clunky, and hot. But as JMR's new SATAStor shows, a RAID array can be compact and quiet.

Imagine a six-drive array of 80GB hard drives that fits entirely in a single 5.25-inch, 1/2 height drive bay. This is made possible by using 2.5-inch drives primarily targeted for laptop systems. We originally saw this JMR SATAStor at the Intel Developer Forum a year ago.

Of course, you can't just drop a six-drive array of SATA drives into your typical desktop system and expect it to work. So 3Ware shipped us an Escalade 9500S-12. This PCI-based RAID controller supports up to twelve SATA drives in a variety of RAID configurations (and in a JBOD—"just a bunch of drives" setup, if that's what you want). The controller offers flexible support for RAID 0, 1, 10, 5 and 50, giving you a choice for the right mix of performance and data integrity.

If you're thinking that the combination of the Escalade and six-drive array isn't really a great fit for a desktop PC, then your guess is correct. It's really well suited to small servers that need to be compact and exist in low-noise environments. Alternatively, it could be used in a workstation for video or audio editing.

We're going to plow right ahead and test it in a standard desktop system anyway—partly just to show we can do it, and partly to see how it compares with internal SATA arrays. We stuck with just a RAID 0 configuration for this article, but we'll be investigating performance in other configurations in the future. This particular hardware setup is somewhat hobbled by the use of 32-bit PCI slots, though the Escalade 9500S is a fully PCI-X-compliant part. It can run in a 32-bit environment, but its real comfort zone is in a system with a 64-bit PCI-X subsystem. The majority of desktop PCs, of course, lack PCI-X capability (not to be confused with PCI Express, which is a completely different beast). At one point, Gigabyte was offering a variant of their 8KNXP motherboard, the GA-8KNXP Ultra 64, with an Intel 875P chipset and a pair of PCI-X slots. But trying to locate one of these is an exercise in frustration. Another possibility is the SuperMicro P4SCT, which seems to be readily available. Both are socket 478 solutions, though, and don't support more recent LGA775 processors.

Alternatively, you could always go for a workstation configuration. A variety of socket 603 (Xeon) and socket 940 (Opteron/socket 940 Athlon 64 FX series) motherboards are available. They're typically pricier (and often larger) than normal ATX boards, which means you'll need a workstation-class chassis.

But we digress. The Escalade controller does work in a 32-bit PCI slot, so we popped it into our storage test bed and took the SATAStor for a spin.

The SATAStor really does pack six 2.5-inch Serial ATA drives into a single, 5.25 inch chassis.

The SATAStor uses 2.5-inch SATA drives. Our unit uses six Fujitsu MHT2080BH 80GB drives with fluid-damped motors. They spin at 5400RPM and have an average access time of 12ms and rated acoustics of 28dB. These tiny drives come complete with an 8MB buffer. Of course, they're not cheap at around $250 apiece. The SATAStor array costs $1,299, so you save some over individual drives.

3Ware offers three different models of the 9500S, with 4, 8, and 12 ports respectively. We tested the drive array with the 9500S-12, which can support up to twelve drives.

We've revamped our test bed so that we can test more recent storage technology. Here's the current configuration:

Component Brand/model
CPU 3.4GHz Pentium 4 Extreme Edition (800MHz FSB); Hyper-Threading enabled
Motherboard and chipset Intel D925CXV, Intel 925X chipset
Memory 2x256MB Kingston DDR2/533 running in dual-channel mode, CAS4-8-8-12 timings
Graphics Nvidia GeForce 6800 GT PCI Express
DVD-ROM Toshiba SD-1502 (check prices)
Audio Sound Blaster Audigy 2 ZS Gamer
Ethernet Intel integrated
Operating System Windows XP Professional, Service Pack 1

Shifting to the 925X platform gives us four native SATA ports capable of supporting NCQ (native command queuing) SATA drives. The ICH6R I/O Controller Hub also supports Intel's own RAID features, which we use in our testing.

We ran four sets of benchmarks:

  • WinBench 99 version 2.0 Disk Tests. We ran the two Winmark tests (Business Disk Winmark and High End Disk Winmark). We also ran the CPU utilization, access-time and transfer-rate inspection tests. The transfer rate was set to maximum throughput.
  • Multimedia Content Creation Winstone 2004, to see how the drives performed under a set of real-world applications, executing scripts that represent real user workloads.
  • Business Winstone 2004, which lets us investigate the impact on a suite of standard desktop applications.
  • PCMark 2004 Disk tests. The new tests use traces from actual disk accesses in key applications areas, including Windows XP boot, applications loading, file copying and general disk usage.

All drives were defragged after every test and the system rebooted. We also ensured that no background tasks were running by executing the command

rundll32.exe advapi32.dll,ProcessIdleTasks

and waiting for the background idle tasks to complete. We also disabled System Restore, remote desktop, and automatic updates.

We ran the tests on two different RAID 0 configurations. The first consisted of a pair of Seagate 7200.7 hard drives (model number ST3160827AS). These are 7200RPM drives that support native command queuing. The two drives were set up as a RAID 0 array with a 128KB stripe size. Second, we connected the JMR drive array to the Escalade controller via six SATA cables on SATA ports 0-5 of the RAID controller. We configured the drives as a single RAID 0 array with a stripe size of 64KB (the controller also supports a 256KB stripe size with six drives). Our net size was about 448GB after overhead.

Later, we configured the drives as a RAID 10 array, which yields a striped, redundant array. The net size was about 228GB after overhead. We didn't run all our benchmarks on this configuration, but it was good to know that it worked.

WinBench 99 2.0 Disk Tests
Although this test is getting quite old, it does reflect application usage. In fact, the WinBench tests were recorded using data collected on how actual applications access the hard drive. Note that the inspection tests, such as transfer rate and CPU utilization are more synthetic.

We set these tests to maximum throughput, which can put a heavy load on the CPU. In real life, you'd rarely see the hard drive hammered in this way. Most of the time, disk accesses occur relatively infrequently or only last for relatively short periods of time.

Although the Fujitsu scored lower on the overall WinBench test, its throughput came close to that of the Seagate-based drive array at the front of the drive (outer tracks). The lower rotational speed and smaller inner track size took its toll on the inside tracks, so inner-track transfer rate was considerably lower. PCI throughput probably played some role in this, though the results were surprisingly good. The really notable score was the CPU utilization—half that of the ICH6R array.

Note, however, that the Fujitsu-based array fared less well in application-based testing. The 64KB stripe size, the slower access time, and PCI throughput probably all contributed to relatively lower scores. But then again, that's not the strong suit for this type of product.

PCMark 2004 Disk Tests
This year's PCMark 2004 disk tests have been substantially beefed up over last year's version. Futuremark has moved away from its own disk tests and is instead using Intel's RankDisk test suite, building its test runs using real usage scenarios.

These tests reflect average desktop PC usage, so it's no surprise that the Seagate array substantially outperforms the JMR array here.

The JMR SATAStor isn't meant to be a primary drive setup for a normal desktop PC. That's clearly reflected in the benchmark scores. However, in a PCI-X based system, we'd expect to see higher overall benchmarks when configured as a RAID 0 array.

The array is also much more flexible than your typical desktop PC array. We configured the array using the Escalade controller's BIOS setup to build a RAID 10 array. Though the overall drive size was about half, because the array is redundant, performance in our initial tests wasn't that much slower. This gives you a fairly fast, redundant array. As a secondary (data) in a system with 64-bit PCI-X slots, you could configure the 3Ware controller to support two of these. Yet, they'd draw far less power and be considerably less noisy than twelve 3.5" hard drives.

The Escalade 9500S-12 is clearly a high-end RAID controller, and it's priced like one, at roughly $700. If your needs are less stringent, 3Ware does manufacture an 8-port version that's under $500 and a 4-port controller for about $300.

In the end, it all comes down to your application needs. We would never recommend the SATAStor for normal desktop use. But for video or audio editing or departmental servers, the 3Ware / JMR configuration might be just the ticket.

Product: JMR SATAStor Compact SATA RAID Array
Web site: www.jmr.com
Pros: Compact, high capacity RAID array; relatively quiet; very easy setup.
Cons: Not a good option for desktop PC users; expensive.
Summary: The SATAStor would be terrific as a secondary drive in a content creation workstation or as a primary drive (using redundancy) in a small server.
Price: $1,299 direct

Product: 3Ware Escalade 9500S-12
Web site: www.3ware.com
Pros: Supports up to 12 SATA drives; easy configuration; PCI-X support; 128MB buffer; low CPU utilization.
Cons: Expensive; needs PCI-X for maximum performance.
Summary: It's much easier to set up than the typical SCSI RAID controller, but 3Ware's offering is not a cheap solution. It does, however, offer RAID 5, 10, and 50 if you need the redundancy and performance. (Note: the 8-port version is $500 and the 4-port version is about $300).
Price: $700 (Est.)

Loyd Case came to computing by way of physical chemistry. He began modestly on a DEC PDP-11 by learning the intricacies of the TROFF text formatter while working on his master's thesis. After a brief, painful stint as an analytical chemist, he took over a laboratory network at Lockheed in the early 80's and never looked back. His first 'real' computer was an HP 1000 RTE-6/VM system.

In 1988, he figured out that building his own PC was vastly more interesting than buying off-the-shelf systems ad he ditched his aging Compaq portable. The Sony 3.5-inch floppy drive from his first homebrew rig is still running today. Since then, he's done some programming, been a systems engineer for Hewlett-Packard, worked in technical marketing in the workstation biz, and even dabbled in 3-D modeling and Web design during the Web's early years.

Loyd was also bitten by the writing bug at a very early age, and even has dim memories of reading his creative efforts to his third grade class. Later, he wrote for various user group magazines, culminating in a near-career ending incident at his employer when a humor-impaired senior manager took exception at one of his more flippant efforts. In 1994, Loyd took on the task of writing the first roundup of PC graphics cards for Computer Gaming World -- the first ever written specifically for computer gamers. A year later, Mike Weksler, then tech editor at Computer Gaming World, twisted his arm and forced him to start writing CGW's tech column. The gaming world -- and Loyd -- has never quite recovered despite repeated efforts to find a normal job. Now he's busy with the whole fatherhood thing, working hard to turn his two daughters into avid gamers. When he doesn't have his head buried inside a PC, he dabbles in downhill skiing, military history and home theater.

Submit a Comment

Loading Comments...