RAID Done Right

Nathan Edwards

Like the eponymous bug spray, RAID gets results. But in this case, the active ingredient isn’t a deadly poison, but hard drives—or, to spell out the acronym, a redundant array of independent (or inexpensive) disks.

RAID represents a storage schematic, a way to use multiple hard drives to accomplish wondrous achievements in automation and capacity. You can chain a number of drives together to create one large super-volume, you can have one drive automatically replicate the contents of another, you can do it all!

So where do you start? With this guide, because while RAID may sound simple, the actual practice of setting up an array is mildly daunting. But before we start attacking the various configuration options that smack you in the face with every RAID setup, we’ll start with the easiest part first, the shopping list.

To set up a RAID, you’ll need at least two items: a motherboard with the ability to create and manage RAID volumes and some hard drives. The exact number of drives will depend on the flavor of RAID you choose, the level of performance you hope to achieve, and your budget, but the drives should be of an identical make and capacity, as your RAID configuration will always be limited by the speed and size of the slowest drive. If you’re planning to string together more than four drives, you’ll likely need to invest in a RAID controller card as well (check your motherboard manual for details about its integrated RAID support).

RAID 0

The RAID variant that offers the fastest speeds and most capacity also comes with the biggest worries

A RAID 0 setup is commonly known as a striped array. Instead of writing all of your data to a single drive, this configuration allows a file to be broken up into smaller chunks, or stripes, which are then written across all the drives in the array. The more drives you add to a RAID 0 config, the faster the overall performance of the array. After all, by adding drives, you’re just spreading out the workload.

To really get the most from RAID 0, you’ll want to play with the stripe sizes. We say play, as there’s no concrete way to gauge what stripe size will be best for your particular setup—short of testing its performance with the apps you’ll be using.
If a file is a pizza, then a stripe is a slice. Slap a 50KB file onto a four-drive array with a 16KB stripe size, and three hard drives will have full 16KB stripes while the fourth will have just 2KB. The sizes affect RAID performance because using smaller stripe sizes often spreads the simultaneous writes and reads across multiple drives, which improves transfer performance for larger files. Using larger stripe sizes allows a single file to be split across fewer disks and, if your RAID controller allows it, will free the unused disks for other access operations. This improves the ability of the drive heads themselves to get to the part of the drive platter with the data.

Hands On

We used four Western Digital Raptor drives in our RAID 0 testing, with a fifth Raptor for the Windows partition. We experimented with stripe sizes ranging from 4KB to 1,024KB, measuring performance with the HD Tach and PCMark05 benchmarks. We achieved the best results with a 128KB stripe size. Using this size, we compared the performance of both a two-drive and four-drive RAID 0 config to that of a single Raptor.

As you’d imagine, the four-drive RAID 0 setup produced the fastest speeds in our benchmark tests. But even striping two drives together gave us a pretty awesome advantage over a single drive. The PCMark05 scores weren’t as much of a blowout as the HD Tach benchmarks, but they nevertheless show that our RAID array is faster than a single drive in every single benchmark the program has to offer.

This immense power, however, comes at a great cost—namely, the safety of the data stored on the array. For if a single drive in your setup fails, that’s it. Your data’s gone. On a two-disk array, striping doubles your chance of data loss due to drive failure. And that risk only increases as you add more drives to the mix.

RAID 0 Benchmarks
HD Tach
PCMark 05
Burst (MB/s)
Average Read (MB/s)
Average Write (MB/s) Score
XP Loading (MB/s)
App. Loading (MB/s) Virus Scanning (Mb/s)
File Writing (Mb/s)
RAID 0 (Four) 414.1 208.7 180.2
11,984.0
23.19
9.07
131.52
272.87
RAID 0 (Two)
358.5
156.2
158.36
8,949.3
15.80
6.07
102.22
266.76
Single Drive
452.1 78.0
102.7
6,329.0
10.42
4.93
77.88
160.51

Two Terabytes? Denied!

Wait! Before you start building a super-array of drives, know that Windows XP does not support partitions greater than two terabytes. It’s just not happening. If you want to, say, chain four terabyte drives together, you’re going to need Windows Vista and a fifth hard drive, because even Vista can’t boot into the partition scheme you’ll need to set up, unless you have an EFI motherboard.

GPT, or the GUID Partition Table, is an updated version of the Master Boot Record partitioning scheme that will let you break through Windows’s 2TB limit on disk sizes. Install the OS on your separate hard drive, then set up your RAID 0 config. When you initialize the disk in Vista’s Computer Management window, make sure you select the GPT partition style instead of MBR.

RAID 1

Making a spare copy of your data will impact performance, but by how much?

Otherwise known as disk mirroring, RAID 1 maximizes protection between two disk drives. Unlike a RAID 0 setup, two drives linked in a mirror configuration don’t double the total capacity of a single new volume. Rather, the capacity of the volume is determined by the size of the smallest drive in the array.

The benefit of mirroring two drives together is obvious; just consider the name of the array. Whenever data is written to a single hard drive, it is instantaneously written to the other drive in the array as well. If one drive fails, you’ll have a copy of all your data. You can then boot off of the survivor by itself or replace your busted drive in the array with a working drive. Your RAID controller will rebuild the array without interrupting normal file operations and return everything to full working order.

This kind of setup is ideal for those who are more concerned about protecting their data than increasing performance. However, don’t misconstrue the benefits of RAID 1 for a data backup solution. A mirrored array is more designed for those, “Oh crap, the hard drive just died randomly” scenarios. A mirrored array won’t protect you from accidental file deletions or malicious software that wipes out your drive (see the sidebar below).

Hands On

Due to the extreme differences between RAID 1 and RAID 0, we expected to see dramatically different results in the relative speeds of the two formats. After all, you’re trading storage speed for sustainability. What we were unsure about was the performance difference between a mirrored setup and a single identical drive in a stand-alone configuration.

As it turns out, the mirrored array actually performed better than a single Raptor. We attribute this to our RAID controller’s ability to select which drive to read data from—it can use one hard drive for one data task, while simultaneously accessing a different data request with the other. Not surprisingly, the mirrored array’s write speeds weren’t as impressive but still bested a single Raptor drive by about 7MB/s.

RAID 1 Benchmarks
HD Tach
PCMark 05
Burst (MB/s)
Average Read (MB/s)
Average Write (MB/s) Score
XP Loading (MB/s)
App. Loading (MB/s) Virus Scanning (Mb/s)
File Writing (Mb/s)
RAID 1 465.9
99.46
109.63
8,085.3
14.90
6.24
82.96
221.53
Single Drive
452.1 78.0
102.7
6,329.0
10.42
4.93
77.88
160.51

RAID 1 As a Backup Solution? No way!

If one drive’s contents are always replicated on another drive in a mirrored RAID configuration, RAID 1 is the perfect backup solution, right? Wrong. Using a mirrored RAID as your de facto backup solution works wonders in certain disastrous occurrences, like if one of your hard drives spontaneously explodes. But RAID 1 doesn’t prevent any of the more malicious (or user-created) data loss issues. If you have a virus on one drive—guess what?—it’s been replicated on the second drive. Or if you accidentally perma-delete a file, it’s gone on both drives. Grab a third-party backup program for your files and let RAID 1 take care of the act of God–type situations.

RAID 1+0

The best of both worlds, the RAID 1+0 combination approach yields great results! Right?

The mix-and-match of RAID setups, RAID 1+0 offers a unique combination of RAID 0 performance with RAID 1 data protection. It’s one of the proud members of the “nested” category of RAID configurations. Like peanut butter on bread, a nested RAID uses one RAID configuration as the base for a second RAID. This hybridization gives you a chance to reap the benefits of both setups, although you’ll never achieve the ultimate benefits of either.

Each RAID controller will handle the setup of the RAID 1+0 config differently, and some won’t even support such fancy storage dreams. Whether the controller defaults to RAID 1+0 or forces you to create an array on top of an array, the basic premise is still the same.

You’ll need at least four hard drives to create a RAID 1+0 setup. Logistically, the drives are split into two pairs. Each pair operates as a mirrored array, or RAID 1. The two pairs are then chained together in a RAID 0 configuration. This gives you data redundancy on the micro level, while still giving you speed and storage benefits on the macro level. You’re safe from data loss as long as you lose only one drive per mirrored pair. You’ll face the same data loss problem that plagues individual RAID 0 arrays if both drives in either of the mirrored pairs suffer an untimely demise.

Hands On

RAID 1+0 performance is far superior to that of a single Raptor drive. But that’s like saying ice cream is tastier than cat food. Two Raptors in a RAID 0 configuration still dominated in the average write portion of our HD Tach benchmarks. And that makes complete sense, as the mirroring portion of the RAID 1+0 array reduces its performance.

RAID 1+0’s average read speeds, on the other hand, are higher than those of two Raptors in RAID 0, but that’s not so much an issue of technology as it is one of scalability. Just for giggles, we fired up four Raptor drives in a striped RAID to get a true, four-drive showdown. The four-drive RAID 0 mercilessly decimated the benchmarks of our RAID 1+0 setup. Average read speeds were 38MB/s faster and average write speeds were 43MB/s faster. If you’re willing to risk catastrophic data loss, RAID 0 is still a speed demon’s friend. But you certainly won’t suffer, speedwise, with a RAID 1+0 array.

RAID 1+0 Benchmarks
HD Tach
PCMark 05
Burst (MB/s)
Average Read (MB/s)
Average Write (MB/s) Score
XP Loading (MB/s)
App. Loading (MB/s) Virus Scanning (Mb/s)
File Writing (Mb/s)
RAID 0 (Four) 414.1 208.7 180.2
11,984.0
23.19
9.07
131.52
272.87
RAID 0 (Two)
358.5
156.2
158.36
8,949.3
15.80
6.07
102.22
266.76
RAID 1+0 334.1
170.73
137.8
10,307.7
22.43
7.90
106.22
225.33
Single Drive
452.1 78.0
102.7
6,329.0
10.42
4.93
77.88
160.51


Make a RAID with Windows

If you want the size benefits of a striped array but don’t feel like setting one up using your motherboard’s controller—or if your motherboard doesn’t include a RAID controller—you can actually create a large, striped drive in Windows itself. Just head over to your computer management screen (you get to it by right-clicking Computer in the Start Menu and selecting Manage). Convert the drives you want to stripe to dynamic disks, then create a new volume and select “striped” for the configuration. Voila! The speeds of the array won’t be nearly as fast as those of a controller-based striped array, but they’re still noticeably faster than a single Raptor’s, with the added benefit that your data isn’t tied to a controller that’s soldered onto your motherboard.

Windows vs Controller RAIDS
HD Tach
PCMark 05
Burst (MB/s)
Average Read (MB/s)
Average Write (MB/s) Score
XP Loading (MB/s)
App. Loading (MB/s) Virus Scanning (Mb/s)
File Writing (Mb/s)
Controller RAID
358.5
156.2
158.36
8,949.3
15.80
6.07
102.22
266.76
Windows RAID
N/A
N/A
N/A
8,107.5
13.6
5.63
100.82
239.74
Single Drive
452.1 78.0
102.7
6,329.0
10.42
4.93
77.88
160.5

Best scores are bolded. HD Tach measures hardware-based volumes and cannot run benchmarks on software-based RAID solutions.

RAID 5

Parity makes a world of difference and barely hurts speeds.

Like RAID 1+0, a RAID 5 configuration is a hybrid combination of data safekeeping and speed. But unlike the former, RAID 5 doesn’t rely on mirroring to preserve your information. It instead uses an alternative method of data redundancy found in RAID setups—parity.

To get into the fine nuances of how parity works would require Excel charts, lots of binary code, and acronyms—lots of acronyms. So we’ll generalize. The mathematics of parity dictates that if you have four drives in an array, the RAID will split each piece of data into three stripes. Each stripe will go to a single hard drive, as it would in a RAID 0 configuration.
The controller then creates a parity stripe based on the three stripes of data. A parity stripe is a logical calculation that allows the controller to re-create any individual stripe that becomes corrupt (or in the case of a drive failure, nonexistent). Similar to mirroring, the lost data is made available to the host machine instantaneously. But the loss of a single drive puts the entire array at risk. Should an additional drive fail—making that two of the four drives dead—all the data on the array is lost. A parity stripe works wonders, just not miracles.

Hands On

RAID 5 gives you the best combination of speed, size, and data savings. Our RAID 5 and RAID 1+0 arrays scored similar speeds, with the RAID 5 squeezing 15 additional MB/s in our HD Tach average read test.

The bonus comes in the fact that our RAID 5 array gave us an additional drive’s worth of space to play around with—450 total gigabytes as opposed to the RAID 1+0’s 300GB of total capacity.

Admittedly, a RAID 1+0 array gives you better data redundancy on paper, but the additional mirroring seems like overkill. In essence, you’d be performing the same maintenance tasks you’d be performing in a RAID 5 array. If a drive goes out in a RAID 1+0, it would be in your best interest to stop what you’re doing and immediately replace the dead drive; the same goes for RAID 5. While the next drive that goes out in your RAID 1+0 array might not be the one to destroy a mirrored pair and consequently your data, do you really want to roll the dice? We wouldn’t, and we’d much rather have the performance and size benefits a RAID 5 array brings.

Why Use a Controller?

All of the benchmarks in this feature were completed using Adaptec’s RAID 31605 controller ($1,000, adaptec.com). In our initial tests, we found that our EVGA 608i chipset-based RAID speeds simply paled in comparison. Thanks to an onboard 800MHz processor and 256MB of DDR2 cache memory, the controller was able to output an average read speed of 211.7MB/s in a simple HD Tach benchmark of a four-drive, striped array. The motherboard-based RAID topped out at 118.9MB/s.

RAID controllers also offer more options and safety features than a motherboard-based chipset, and the motherboard RAID itself is limited to the number of free SATA ports you have. In contrast, our controller supports up to 16 SATA drives.

Motherboard RAID vs Controller RAID
HD Tach
PCMark 05
Burst (MB/s)
Average Read (MB/s)
CPU Use (MB/s)
Score
XP Loading (MB/s)
App. Loading (MB/s) Virus Scanning (Mb/s)
File Writing (Mb/s)
Motherboard 234.5
118.9
2.0
10,525.0
25.54 12.61
82.53
129.05
Controller
473.3
211.7
2.0
12,162.0 23.76
9.08
132.57
282.73

Best scores are bolded. Arrays were tested using a four-drive RAID 0 configuration.

Around the web

by CPMStar (Sponsored) Free to play

Comments