In the November 2007 issue, we took an in-depth look at RAID—short for Redundant Array of Inexpensive (or Independent) Disks—and broke down the pros, cons, and most importantly, speeds of the various RAID permutations you would find on a typical multidrive setup. Here we’ll examine the medium itself: the RAID controller, which tells the drives in a RAID setup how to interact. As you’ll see, there are RAID controllers of differing types, technologies, and price points, and we want to learn whether these variations translate into performance differences. After all, even the fastest RAID configuration ultimately depends on the performance capabilities of its physical host.
A RAID array can be set up in one of two ways: You can use the controller that’s built into your motherboard’s chipset (if it includes one), which requires you to hook up your drives as normal and then edit a few BIOS settings, or purchase an external controller card, which boots its configuration menu before your operating system even loads.
Our mission is to test the performance of RAID setups using both low- and high-end RAID cards from five different manufacturers and compare those results against the performance of two common motherboard solutions. Once the dust settles, you’ll know which RAID controller will give you top performance and exactly what features you get for your buck!
Every aspect of our experiment had to be considered in order to ensure meaningful results.
To test the performance of motherboard-based RAID controllers we turned to Nvidia’s nForce 680i chipset and Intel’s P35 chipset. The former represents the typical testing environment we use for all our storage benchmarking and is a solid example of what you’d find on a high-end motherboard. The Intel-based board provides a fair representation of RAID performance on a midrange machine. Given Intel’s dominance in the performance-computing market right now, we didn’t bother testing an AMD-compatible motherboard. What’s the point?
For our controller cards, we picked a combination of host-based and discrete models: With the former, the controller uses the rig’s processor to handle RAID functionality (just as a motherboard chipset does), while discrete controllers remove your computer from the equation by supporting a low-powered processor right on the card. As you might expect, host-based controllers are cheaper than their discrete counterparts. Besides selecting controllers for both card types, we also sought to vary the price points in each category.
All the testing, save that done on the MSI P35 Neo2-FR motherboard, was conducted using our standard storage benchmarking system: an Intel Q6700 on an EVGA nForce 680i motherboard, a single EVGA GeForce 8800 GTX videocard, and a single 500GB Western Digital Caviar SE16 hosting the Windows XP operating system. Each RAID test uses four Western Digital 150GB Raptor drives.
We elected to use two RAID configurations for benchmarking: RAID 0 (aka striped) and RAID 5. This allowed us to study the price/performance equation from two different mindsets: speed and redundancy. As we noted in our previous RAID feature, you won’t find a faster storage configuration than RAID 0, in which the drives serve as one volume and your data is written and read across them concurrently. A RAID 5 setup balances the performance gains of striped arrays yet still offers a degree of data redundancy.
We’re using the same standard suite of benchmarks we normally use for storage testing, with a few changes. We’ve downgraded our HD Tach program to version 220.127.116.11 to ensure we can run both read and write tests on the arrays. This synthetic benchmark bypasses any software on a machine to get right to the drives themselves, measuring the subject’s
For a more real-world test, we’re using PCMark05. We’ll compare the scores from the program’s subset tests—XP Startup, Application Loading, General Usage, Virus Scanning, and File Writing—as well as the overall score given to each test subject.
Finally, we’ve created our own real-world benchmark to further test these arrays under a typical usage scenario. We time how long it takes to write a 1080p multimedia slideshow to the array using Photodex’s ProShow, based on files read from the array, while simultaneously using Adobe Premiere to convert a DVD rip on the array into an uncompressed AVI file. As these tasks also maximize our processor’s usage, it allows us to discern the potential performance pitfalls of a card that uses said processor for its RAID calculations.
EVGA nForce 680i
This motherboard, chosen for the 2007 Maximum PC Dream Machine, comes with support for up to six SATA devices. But it’s not just the potential size of the RAID that gets us going. The motherboard also provides a decent list of configurations to choose from: RAID 0, 1, 0+1, 5, and JBOD (just a bunch of disks). It’s a surprisingly generous offering, given the fact that some of the add-in cards we’re testing lack such variety.
Setting up the RAID itself is a breeze—you switch the SATA ports to RAID in the BIOS and then use a handy utility to select the drives, array type, and configuration options you want.
MSI P35 Neo2-FR
MSI’s P35 Neo2-FR is a respectable midrange motherboard that sports the flashin’ new Intel P35 chipset with an ICH9R south bridge. We expect RAID performance with this chipset to rival that of EVGA’s nForce platform, which is an older chipset stuffed onto a more high-end motherboard.
You get only five SATA ports on the P35 Neo2-FR: Four operate under the ICH9R chipset, with one running on an integrated Marvell 88SE6111 controller. This could have some bearing on performance when compared to the EVGA nForce 680i, which uses the south bridge for all six SATA ports.
HighPoint RocketRAID 2300
HighPoint’s entry-level RAID controller card is but a mere PCI Express x1 model—though in many ways, this makes it ideal for a typical motherboard setup. If you have the available connector on your motherboard, you won’t have to burn one of your x8 or x16 slots for the card.
The host-based controller supports five different RAID levels: 0, 1, 5, 1+0, and JBOD. The controller features four ports, allowing you to connect up to four SATA drives using standard SATA cables.
We’re mildly concerned that the controller’s lack of any onboard memory might hurt the card’s performance. But that’s more an issue of the card’s price point—you aren’t going to see memory packed onto entry-level products.
HighPoint RocketRAID 3510
One of the most expensive cards in HighPoint’s line of RAID controllers, the RocketRAID 3510 trumps its lesser cousin, the RocketRAID 2300, with a few key features. You’ll also find a multilane connector on the 3510 rather than a four-port SATA connector, although you get the same effect: The included breakaway cable supports the same number of drives—four.
More importantly, the 3510 controller comes with a built-in Intel IOP 81341 processor. The 800MHz proc pulls RAID actions off your rig’s CPU. We expect this critical difference to shine in our more taxing RAID benchmarks, as we’ll want every bit of our CPU’s power going to the rendering test, not the RAID functionality.
The card also comes with 256MB of onboard DDR2 memory and is fashioned for a PCI Express x8 connection.
Adaptec’s entry-level 1430SA card supports only four RAID levels: 0, 1, 1+0, and JBOD. Critically missing is any support for a RAID 5 configuration—something all the other entry-level cards we’re testing support. While this automatically disqualifies Adaptec’s card from the RAID 5 portion of our comparative benchmarks, it doesn’t necessarily bump the card to last place. The 1430SA might very well offer the best performance for its price in a RAID 0 configuration—which could be fine for folks interested in only RAID 0, or one of this card’s other configs.
The PCI Express x4-based card comes with no extra frills. There’s no onboard memory, nor is there a dedicated processor on the card to handle RAID requests. You can connect up to four SATA drives to the array via four standard SATA ports.
If RAID controllers were cars, you’d find Adaptec’s 5405 on a luxury lot. This little device comes packed with every feature, accessory, and upgrade an enthusiast could ask for.
The 5405 sports an onboard 1.2GHz dual-core processor to handle RAID functionality—that’s more CPU than this article’s author has on his laptop. We’re expecting this, as well as the card’s 256MB DDR2 cache, to spit out massive performance. But we’re also curious to see how a card this stacked will benchmark
The 5405 supports an obscene number of RAID levels (0, 1, 1E, 5, 5EE, 6, 1+0, 50, and 60), making it the clear winner in the options category. The card uses a PCI Express x8 interface for the connection mechanism and a multilane port for SATA connections via a breakaway cable.
AMCC 3ware 9650SE-4LPML
The 9650SE-4LPML represents AMCC’s entry-level RAID controller. It runs off a PCI Express x4 interface and supports a decent array of RAID configuration options: 0, 1, 5, 1+0, and JBOD.
The card boasts an onboard 266MHz PowerPC CPU, which does its part to keep your general computer operations separate from the RAID equation. However, we ask ourselves: Is 266MHz enough? In the grand scheme of things, especially given that our test machine runs a quad-core Q6700 processor, will such a relatively underpowered onboard chip have much of an impact on our benchmarks?
The controller also comes with 256MB of onboard DDR2 memory. SATA drives are connected via a single multilane port using an included breakaway cable.
AMCC 3ware 9650SE-24M8
The high-end version of AMCC’s 9650SE RAID controller is a bit bulkier than the 4LPML card—and that’s due to its extra connection ports. You get three multilane connectors on the large 24M8 card—for a maximum of 24 drives, with the appropriate breakaway cables, of course. The controller attaches to your motherboard using a PCI Express x8 connection.
This card expands upon the 4LPML’s RAID options in a rather dramatic fashion by supporting 0, 1, 5, 6, 1+0, 50, and JBOD. That puts it nearly on par with Adaptec’s 5405 controller.
We’re mildly taken aback by this pricey card’s paltry power potential, as the 24M8 runs a 266MHz processor, just like its far less expensive cousin. While this controller has double the RAM (512MB) of the 4LPML, we’re skeptical that this will provide a huge performance boost.
LSI Logic MegaRAID SAS 8208ELP
On paper, LSI Logic’s MegaRAID SAS 8208ELP controller looks like a respectable contender at the entry-level end of the RAID marketplace. The controller doesn’t use an onboard processor to shuffle RAID duties from your CPU to the card, but it remains to be seen if this will have a demonstrable effect on next-generation computing systems.
The card does come with two separate multilane connectors for adding up to eight SATA devices with the breakaway cables that come included. You can then slap these drives into one of four RAID configurations: 0, 1, 5, or 1+0.
In practice, we found this PCI Express x4-based card to be quite frustrating, primarily because the drivers LSI shipped with the card flat-out didn’t work. A quick run to LSI’s website for updated drivers fixed the problem, but the fact remains: Out of the box, the 8208ELP does not function.
LSI Logic MegaRAID SAS 8888ELP
Thankfully, the 8888ELP works straight out of the box with the drivers LSI provides. As well it should—this is the company’s high-end RAID product, designed as a discrete-performing controller for handling RAID 1, 5, 6, 1+0, 50, and 60 arrays. The controller itself comes with the same two internal multilane connectors as the 8208ELP, with an additional two external connectors located on the rear of the card.
The 8888ELP attaches to your machine using a PCI Express x8 connector. The CPU itself chugs along at 500MHz. This, plus the controller’s 256MB of DDR2 memory, should help the card hold its own in our RAID benchmark showdown. It will also give us an interesting comparison point: What’s more important, a faster processor or more memory?
Yes, there's something to be said for add-in cards, but a costlier controller doesn't necessarily pay off.
It’s only fair that we start our comparisons by looking at the host-based cards and motherboards. The relatively low cost of entry in these two categories makes this the natural starting point for a discussion about how controllers can impact the performance of a RAID. We had high hopes for our motherboards going into the slugfest. After all, these two products each represent pinnacles of performance: EVGA’s nForce-based motherboard is a tried-and-true favorite in our Lab, whereas the MSI board sports one of Intel’s newest chipsets. But even with all that respective might behind them, the two contenders were no match for Adaptec’s 1430SA host-based add-in card. In two of the three official RAID 0 benchmarks, the 1430SA overtook the speediest of our motherboards, MSI’s P35 Neo2-FR.
The only anomaly was a stunning upset by MSI’s P35 Neo2-FR in our real-world encoding benchmark. We were surprised to see the board perform so dramatically better than any other contender in the host-based category. We have two possible explanations: Just considering the two motherboards, the EVGA model uses a single south bridge to control the functions of six SATA ports. The MSI board shuffles a five-port load across two separate onboard controllers. Then there’s the simple possibility that this benchmark performs better on an Intel platform—we’ve seen it happen before and are apt to believe it to be the case this time, given the close performance of the two boards in the two other benchmarks.
Concerning RAID 5 performance, we must first note that the Adaptec 1430SA card was excluded due to its lack of support for RAID 5. It’s a shame, too; HighPoint’s RocketRAID 2300 squeaks out the performance win, but it’s not an outright domination. Given how well the Adaptec card performed on our RAID 0 testing, we believe it would have stood a good chance of taking the crown in the RAID 5 benchmarks as well.
We found that both motherboards had a great deal of trouble with write performance in RAID 5. This killed their scores for all tests across the board, as both of our real-world benchmarks depend on a storage device’s read and write capabilities. This leaves the RocketRAID 2300 as the default winner. In no way would we want to suffer through the abysmal write times of the two motherboards for any length of time.
It was interesting to see virtually no variance whatsoever in CPU usage, regardless of whether our array was being fielded by the motherboard or the host-based controller card. We remain unconvinced that a host-based controller’s performance is hurt in a RAID setup merely because it relies on the CPU—at least, if you’re running a multicore processor. The particulars of the card and the RAID configuration are what ultimately matter—as became apparent during the discrete portion of our testing.
If you’re willing to spend the extra dough, the best discrete RAID controllers punish the competition to satisfyingly bloody results—but there are still duds to be found in this category. Both cards that used high-powered processors (800MHz or higher) destroyed the RAID 0 benchmark numbers of their host-based counterparts. While the ultimate winner was Adaptec’s 5405 controller, HighPoint’s RocketRAID 3510 achieved excellent results in our RAID 0 HD Tach and PCMark05 benchmarks. The real-world significance of the card’s power was less apparent in our conversion benchmark, where the RocketRAID 3510 card trumped only the next-best contender by 11 seconds.
By contrast, Adaptec’s 5405 card shined brightly on every individual test we put it to. The card topped all others in our HD Tach read test and utterly blew away the second-place controller by nearly 100MB/s on the write test. The 5405 pulled in awesome numbers in PCMark05 and also did well in our real-world benchmark. This ultimately amounted to a savings of about 50 seconds compared to the RocketRAID 2300, but compared to the nForce chipset, we’re talking about a difference of six minutes.
We were most surprised to see the relatively lukewarm performance of the absurdly expensive 3ware 9650SE-24M8 controller in all of our benchmarks. If anything, this proves that you can’t just toss money at a RAID controller and expect dynamite performance. For a $1,600 (MSRP) contribution to AMCC, you get RAID 0 performance that’s no better than what AMCC’s $400 model is capable of, or even than the $150 entry-level HighPoint RocketRAID 2300 controller. Yikes!
Adaptec’s 5405 controller and HighPoint’s RocketRAID 3510 traded shots during our RAID 5 testing. The former gave us the highest write speeds of any RAID card we tested in this showdown, leading to a squeaks-by-at-the-finish-line victory in our overall PCMark05 benchmark. However, the RocketRAID 3510 had consistently excellent performance across all read-related tests, overtaking Adaptec’s 5405 controller by 14 seconds in our encoding benchmark.
The other cards we tested paled in comparison: 3ware’s high-priced offerings gave us decent RAID 5 performance as compared to, say, a motherboard. They nevertheless showed little variation among the two price points. Adding insult to injury, both cards were topped by the host-based (and far cheaper) HighPoint RocketRAID 2300.
As for LSI’s cards, well, we were amazed to see such an absolute nosedive from both the 8208ELP and 8888ELP controllers. Just to make sure we weren’t flubbing something, we tested these cards in both RAID 0 and RAID 5 arrays on both of the motherboards mentioned in this feature. No dice. Both LSI cards turned in abysmal performances in RAID 5—in fact, their complete inability to perform to reasonable expectations constitutes a failure in our book. And while the 8888ELP functioned in RAID 0, its scores were average at best. We suspect this might be a driver issue of some sort, given the problems we had installing drivers to begin with on the 8208ELP card. However, we can’t argue with the numbers.
You have all the data, but we're taking it one step further: Here's what you should buy!
At the end of the day, the RAID equation is pretty simple. You’ve seen the numbers and the analysis, so let’s take a big-picture approach before you rush out and plunk down money at your retailer of choice. The two motherboard-based RAID controllers we tested are perfectly acceptable solutions if you want a little speed boost in a RAID 0 environment, especially since it costs you nothing extra. In fact, depending on the motherboard you have, you may very well see a measurable difference between that and an entry-level host-based card.
If you’re planning to run RAID 5, ignore everything we just said. You will want to have an add-in controller because if the two products we tested are any indication, motherboards just up and die when it comes to RAID 5 performance. You can pick up a host-based controller on the cheap and see a performance boost—provided the card supports RAID 5. In this respect, we have only the RocketRAID 2300 to go by, and based on our results, we think this relatively inexpensive host-based controller card earns its keep. But don’t assume that you’ll get comparable performance from a similarly spec’d/priced alternative. As this roundup has shown, RAID-controller performance varies wildly. You might luck out and score a decently performing product, but you might also select a card that just falls to pieces on your configuration of choice.
The surest way to avoid any pitfalls, especially if you’re planning on testing out multiple RAID configurations, is to pick up a good discrete card. Don’t go after the most expensive one you can find. As we’ve learned here, performance does not necessarily scale with price.
That said, we found great success using Adaptec’s 5405 card in a RAID 0 environment. It was the fastest card in all of the benchmarks, and it offers substantial performance benefits over a similarly configured motherboard RAID environment. You can connect a bevy of hard drives to the card, and it comes with the most bountiful RAID options of any device here. If you don’t mind an excruciatingly long initialization period, this card is perfect for those who want to dabble.
If you’re just looking at RAID 5 performance, however, we have to tip our hat to HighPoint’s RocketRAID 3510. It’s less expensive than Adaptec’s 5405, albeit also less feature-packed. But when it comes time to throw down, the card squeaks out the performance win over Adaptec’s 5405. The 5405 beats the RocketRAID 3510 in RAID 5 write speeds. But it’s still not enough to overtake the Rocket Raid’s speedy reads.
The numbers say it all: We saw only a 14-second difference between the two cards in our real-world encoding test. If you can stomach that loss and want a card with a ton of RAID options, get Adaptec’s 5405. If you just want RAID 5 performance and a card that can also hold its own in RAID 0, opt for the cheaper RocketRAID 3510.
But whatever you do, avoid LSI cards. Two out of two failures across two separate motherboards and a flurry of different hard drives does not a happy storage enthusiast make. Don’t say we didn’t warn you!