RAID Level 0 provides the speed you need for handling large graphics and video files.
Kelli Wiseth
RAID -- the word may conjure up images of late-night gangster movies and Eliot Ness storming into secret hideaways. But although this storage technology doesn't involve moonshine, it may quench your thirst for speed in graphics, video, and prepress environments. RAID systems provide faster access to data than do regular, stand-alone hard drives -- you'll especially appreciate the increased speed if you work with large files.
A RAID (redundant array of inexpensive disks) system combines multiple hard drives so that they behave as a single unit. On your Mac's desktop, you see a single volume, but behind the scenes may be anywhere from two to seven hard drives working in unison. It's the old divide-and-conquer approach: Files are split among the drives, as is the workload -- the reads and writes occur on multiple physical drives simultaneously, resulting in faster speed than when you use a single hard drive.
There are six levels of RAID -- 0 through 5 -- for dealing with multiple drives. Five of these levels provide various degrees of reliability, or fault tolerance (see the "Explaining RAID" sidebar). RAID 0 has no built-in reliability features, but it's the fastest level -- and it's the focus of our testing, since speed is of the essence for publishing, prepress, and video applications.
For this report, we tested 17 RAID 0 systems -- combinations of hard drives, RAID software, and interface cards -- in capacities of at least 4 GB. Fourteen vendors are represented, including such newcomers to the Mac market as DPT and xiStor. A veteran of the PC marketplace, DPT is offering its complete line of storage products in Mac versions; xiStor is a recent startup that sells only Mac products. Eight of the systems we tested are for the ultrafast PCI bus, which was recently introduced to Macs in the new Power Mac line; eight are for NuBus Macs; and one, the La Cie Joule RAID, connects to the Mac's built-in SCSI bus.
Among the vendors that are noticeably absent are APS, MegaDrive, Micropolis, and Storage Dimensions. They were in the process of upgrading their RAID systems at the time we did our testing. Some vendors that use products from Adaptec (a leading vendor of SCSI controllers for PCs) did not have systems available at test time, because Adaptec was acquiring Trillium Research, maker of the Remus RAID software (included with many of the products we tested). Trillium Research will become a wholly owned subsidiary of Adaptec and will continue to sell its Remus software to other vendors.
Although we tested only complete systems, do-it-yourselfers can buy drives, SCSI-2 interface cards, and software and build RAID arrays from scratch. RAID software packages such as CharisMac's Professional Storage Suite ($199 list for RAID 0 and 1), Trillium's Remus ($590 list for RAID 0, 1, 4, and 5), and FWB's RAID ToolKit ($399 list for RAID 0 and 1) are the brains behind the software-based RAID implementations we tested.
Only one system -- the DPT SmartRAID -- is a hardware-based RAID implementation. The other systems include Mac-based RAID software (such as the programs just mentioned), but the SmartRAID includes a PCI SCSI-2 interface card that contains a RAID controller and a 32-bit 40-MHz processor. You'd think that with all the heavy-duty hardware, the SmartRAID would blaze past the software-based RAID systems. But we were disappointed by the SmartRAID's speed -- the software-based RAID 0 products beat the SmartRAID hands down in our tests.
The Case of the Non-RAID RAID
So just what is a redundant array? In the late '80s, researchers at UC Berkeley were looking for ways around the high cost of the high-capacity hard drives used for minicomputers. The scheme they originally proposed (and documented in a paper with a title reminiscent of a Nancy Drew mystery: "A Case for Redundant Arrays of Inexpensive Disks") consisted of five ways -- RAID 1 through 5 -- of combining what were at the time relatively low-cost drives (which had lower reliability ratings as well). Since these drives weren't built to the rock-solid specifications of the high-priced storage used in large-systems environments, a key component of the RAID scheme was built-in fault tolerance -- data redundancy -- which enabled rebuilding of the information on any of the drives in the array if a drive failed. Until recently, RAID has typically been cost-justifiable for personal computers only in the LAN market, for providing fault tolerance for servers. Interest in RAID for nonserver Macs is being stimulated by such factors as promises of overall speed improvements, particularly for high-end graphics. Due to falling hard-drive prices, new PCI hardware development, and the influx of PCI-card vendors to the Mac market, RAID is fast becoming more viable -- read cost-effective -- for Macs. The RAID systems we tested start at less than $3,000 -- as low as $2,799 for the MacProducts Magic 4 GB RAID.
Although RAID 0 wasn't part of the original Berkeley scheme, it is like the other RAID levels in that it "stripes" data across multiple hard drives. The particulars of striping vary among the RAID levels, and in all cases but RAID 0, either a duplicate of your data or information about that data is also saved to disk -- this is the redundant aspect of RAID. Stripes contain the number of blocks of data that can be written (or read) across the entire array; the default stripe size for the products in this review is 64K.
You can see how the speed improvement occurs with RAID if you think about the notion of striping. For example, whereas a single 4-GB drive writes a 64K block of data on a single disk, a 4-GB array composed of four 1-GB drives can simultaneously write four 16K chunks. (The more drives in a RAID array, the faster it is.) When reading files back, splitting the reads among the four physical drives also results in overall faster delivery. RAID 0 doesn't have the overhead of calculating redundant data, so it's the fastest of all RAID levels. Since RAID 0's raison d'etre is providing fast data throughput, speed is the most important criterion. We tested each system, focusing mostly on how it performed with high-end graphics applications such as Photoshop and QuarkXPress. We also ran key MacBench 2.0 test suites and a Finder copy (see the "Need for Speed" charts).
Numbers Racket
In addition to revealing the fastest systems, our tests shed light on several factors you should take into account before buying. Without a doubt, PCI provides faster throughput than NuBus, as you can see from our results. For example, we tested both a PCI and a NuBus version of the FWB SledgeHammer 4100 FMF-W (according to FWB, the only difference between the two systems is the interface). Opening a 32-MB Photoshop file took about 38 percent longer on the NuBus system -- 72 seconds compared to the PCI version's 52 seconds. You can expect faster results with Photoshop 3.0.4, which wasn't available during testing but is now shipping.
In a production environment, where you may be saving a file every 15 minutes, you'd save about 10 minutes over the course of an 8-hour day, which adds up to a full 40-hour week a year. Although that may not seem like much, consider all the other tasks you may do in Photoshop or any other disk-intensive application -- rotating, applying filters, resizing -- and your time savings can easily amount to weeks.
Regardless of whether your Mac is NuBus- or PCI-based, make sure the mix of components makes sense: There's little reason to invest in a RAID 0 system if you don't have a SCSI-2 Fast and Wide interface and Fast and Wide drives. This isn't rocket science -- it's basic math: A Power Mac's SCSI bus can handle only up to 5 megabytes per second, so attaching a Fast and Wide drive capable of 20-megabyte-per-second throughput is a waste -- like slamming your foot on the brakes just as you merge onto the autobahn.
A case in point: Compare the results for the La Cie Joule RAID with those for our baseline storage system, a non-RAID Apple 1080 external drive. Because the Joule RAID also connects to an external SCSI bus and is thus limited to 5 megabytes per second, the SCSI bus becomes the bottleneck, making the Joule RAID fare little better than our baseline drive. At $3,999 for 8 GB of storage, the Joule RAID may sound like a bargain, but you'll be disappointed with its speed.
Hard-drive mechanisms should also support some type of SCSI-2 bus implementation -- Fast, Wide, or Fast and Wide. Each of these has a different bus width and speed. So although the spec for SCSI-2 is 5 megabytes per second, a Fast implementation supports up to 10 megabytes per second; add Wide to this (for 16-bit width instead of 8-bit), and you should see speed increases. With the exception of three systems -- the DPT SmartRAID, the La Cie Joule RAID, and the xiStor xi.RAID -- the hard-drive mechanisms in the arrays we tested support SCSI-2 Fast and Wide. DPT's mechanisms support Fast but not Wide SCSI-2, La Cie's are narrow Quantum mechanisms, and xiStor's are narrow Seagates.
Also be sure the tasks you perform and the applications you use are disk-intensive. In our QuarkXPress test -- saving a document as an EPS file -- speeds of the RAID systems were either the same as or only 1 percent better than those for our baseline Apple 1080 drive. Since this task involves converting data into PostScript code, it is dependent on the speed of your CPU rather than that of the drives. In cases such as this, where the Mac's processor does most of the work, the money you'd spend for a RAID system would do more good if you upgraded your Mac.
Two-Timing Buses
Two-timers in old gangster movies usually ended up wearing cement shoes, but two-timing (as in doubling up) on the SCSI can really help your system get the lead out. When it comes to SCSI buses and RAID, more is better. Vendors such as ATTO and MicroNet sell (independently as well as part of their systems) dual-channel SCSI-2 interface cards in NuBus as well as PCI configurations. A dual-channel SCSI interface card contains two SCSI controller chips and ports, thus providing for up to 15 devices on each of two separate buses. The advantage of using a single dual-channel card is that you lose only one NuBus or PCI slot in your Mac. Conley ships the MicroNet Raven Pro (a dual-channel card) with its NuBus RAID system, but the system uses only a single channel because of the NuBus bottleneck.
Another approach some vendors take is to bundle two separate SCSI-2 interface cards in the package. The Optima DisKovery 4200W and the Microtech BLUE Streak 3400 each come with two SCSI-2 Fast and Wide interface cards. In terms of pricing, you may pay a slight premium for the slot-saving convenience of a single dual-channel card. For example, the ATTO ExpressPCI comes in both single- and dual-channel versions; the single-channel ATTO ExpressPCI costs $395, and the dual-channel version is $895. In terms of speed, either a dual-channel card or two single-channel cards will give you better speed than one single-channel card.
In a single-channel RAID 0 system, the physical hard drives are daisy-chained to each other and in turn to the SCSI bus. The dual-channel approach connects one half of the array -- which may be composed of multiple physical drives -- to one channel and the other half of the array to the other channel, whether the channels are on the same card or on two separate ones.
A dual-channel RAID 0 system is faster, because data is passing through two channels instead of one. In our tests, the NuBus Optima DisKovery 4200W, with its two ATTO SiliconExpress IV cards, was faster on average for the suite of Photoshop, QuarkXPress, and Finder-copy tests than the CharisMac Anubis 4 GB RAID, a PCI single-channel system. If you look at just the results of the Photoshop tests (the most important ones, given the focus of RAID 0), you see that the top five performers overall -- the MicroNet DataDock Wide Storage System, Microtech BLUE Streak 3400, MacProducts Magic 4 GB RAID, DGR Ultrastar 4.0, and Optima DisKovery 4200W -- were dual-channel systems. Of these, the NuBus Optima system was even slightly faster than the FWB SledgeHammer 4100 FMF-W with a single-channel JackHammer PCI card.
The power of dual-channel systems is exemplified in our 1,024K-sequential-write test. The top three performers -- the MicroNet DataDock Wide Storage System, the Microtech BLUE Streak 3400, and the Optima DisKovery 4200W, which are dual-channel systems -- were about four times as fast as our baseline Apple 1080 drive. On the other hand, the single-channel NuBus FWB SledgeHammer 4100 FMF-W and xiStor xi.RAID were only one and a half times as fast.
Overall speed is influenced a great deal by the type of files you normally work with. For example, the DPT SmartRAID has been optimized to work with small files, and as a result, it fared dismally in our tests involving a 32-MB Photoshop file. The SmartRAID did exceptionally well, however, in the MacBench 2.0 Disk Mix test, a test of key business-productivity applications that is optimized for small, "bursty" reads and writes. Although the Disk Mix test is not a network test per se, the type of read and write activity it models is comparable to that of a file server -- bursty reads and writes, typically with small file sizes. We'd be curious to see how the DPT SmartRAID performs in a network environment, but we don't recommend it for desktop publishing or prepress work -- connected to a Power Mac 9500/132, it fared worse than the baseline Apple 1080 drive on the Power Mac 8100/80 in all tests except the Disk Mix test.
It's a Setup
Installing and configuring any RAID system takes some time. In addition to the physical installation of the hard drives, the interface cards, and the connections between them, several layers of software are involved, including drivers for the SCSI interface card, drivers for the hard drives, RAID software for striping and configuring the drives, and sometimes management software. Despite this, we had little difficulty installing and configuring most of the systems -- although the process was sometimes time-consuming.
The software bundled with all these systems (except the xiStor xi.RAID, for which RAID 1 software costs extra) lets you configure them for at least RAID 0 and 1. Most of the products ship with Trillium Research's easy-to-use Remus Lite, which lets you configure an array for RAID 0 or 1. With the Remus Monitor control panel (and the Remus Responder installed on networked Macs), administrators are able to keep a watchful eye on the status of all the Remus-configured RAID systems.
In addition to striping and mirroring for RAID 0 and 1, Remus lets you configure an array to span multiple disks, which consolidates multiple volumes into a single volume, for easier management. Spanning can be a real boon in a graphics environment or for any desktop littered with numerous, variously sized hard drives. Note that you won't see the kind of speed improvement you get when you stripe an array, since with a spanning configuration, writing to the disks that make up the volume happens sequentially rather than simultaneously.
The DGR Ultrastar 4.0, MacProducts Magic 4 GB RAID, MaxConcept MaxRAID SW4100e, Mirror RAID, and Spin Whirlwind are bundled with Remus Lite. The La Cie array also uses Remus Lite, but the software is bundled under the name La Cie Joule RAID. For $295, you can upgrade to the full version of Remus, which gives you RAID 4 and 5 support in addition to support for RAID 0 and 1. Remus' documentation is well thought out and includes tutorials on building various RAID configurations.
CharisMac's Anubis RAID software has the most user-friendly interface -- it's simple enough that you don't need to look at the manual. Anubis RAID, which supports RAID 0 and 1, comes with the CharisMac Anubis 4GB RAID and the Microtech BLUE Streak 3400. The Conley systems ship with Conley's SoftRAID -- the same RAID software that MicroNet bundles with its RAID systems and that Apple bundles with its Workgroup Servers under their own respective names. It's also easy to use, and the documentation is excellent. SoftRAID supports RAID 0 and 1.
At the other end of the spectrum is the DPT Storage Manager software, which proved to be a disappointment. We were teased by references in several places in the manual to some great features -- remote management over both a network and a dial-in connection, for example -- only to find that these features were yet to be implemented in the Mac version. The PC version provides numerous low-level configuration settings that can be adjusted to improve throughput, but regrettably, these settings are also not implemented in the Mac version. And although DPT Storage Manager supports RAID 0, 1, and 5 for the Mac, you have to go elsewhere for driver software -- it's not included with the DPT system. At DPT's suggestion, we used La Cie's Silverlining, but La Cie confirmed that it does not have a bundling agreement with DPT.
Insurance Game
Once you've configured your system, you shouldn't have to use the RAID software again, unless you need to add to the existing array or change the configuration. And pray that you don't decide to do that with a RAID 0 array, because if you do, you'll have to back up all your data to tape or some other medium and then reconfigure the whole system. You should back up a RAID 0 system regularly anyway -- remember that unlike other RAID levels, which are fault-tolerant and enable you to reconstruct your data in the event of a disk failure, RAID 0 does not include built-in recovery features.
You should back up even fault-tolerant RAID systems as well. Although it might seem "redundantly redundant" to back up a RAID 1 mirrored system or a RAID 5 array, it isn't. Fault tolerance means that the array can recover from damage to one of the physical disks in the array by reconstructing files from the redundant data stored across the remaining disks. If two (or more) disks fail, you're out of luck. Fires, floods, or other acts of nature -- a RAID system won't help you overcome these.
Case Closed
The need for speed for DTP, prepress, and video production never diminishes, and using RAID 0 is being touted as one way to increase it. But our tests demonstrated that getting a RAID system is worth the cost only in certain situations. If you've got a PCI-based Power Mac, you'll see a great speed boost when you work with Photoshop files, for example. Any RAID system you get should have hard drives that support Fast and Wide SCSI-2, or you'll be sorely disappointed. Also, RAID systems built around a dual-channel SCSI-2 Fast and Wide card, or two single-channel cards, have an edge over those that use a single channel.
The MicroNet DataDock Wide Storage System is our PCI pick. Its compact enclosure can hold two hot-swappable drives, so you can move the drives to other DataDocks at other workstations. In addition, if you configure the DataDock for RAID 1, the hot-swappable feature will be helpful if one of the drives fails. Better yet, MicroNet offers tech support 24 hours a day, seven days a week.
Although not the fastest in our tests, the DGR Ultrastar 4.0 and MacProducts Magic 4 GB RAID are good buys too. Each took about 40 percent less time than the baseline Apple drive to perform our Photoshop tests. In addition, each company provides a toll-free tech-support number and a five-year warranty.
Among the NuBus systems, the Optima DisKovery 4200W gets the nod. Its two ATTO SiliconExpress IV Fast and Wide interface cards push it a nose ahead of the rest of the pack. Although Optima's standard warranty is only one year, you can extend this to two or three years at an additional cost. And Optima provides cross-ship replacement, meaning that it will send you a replacement unit while you are returning a bum drive. In addition, in case you decide in the future that you need fault tolerance, Optima's software supports RAID 5 (if you have the drives for it). Still, considering how much faster PCI systems are than NuBus ones, graphics and video professionals still working with NuBus Macs might be better off upgrading to a PCI Mac and taking a RAIDcheck.
Kelli Wiseth is a MacUser contributing editor. Senior project leader Kristina De Nike managed the testing for this report.
The Bottom Line
TIME IS MONEY -- and the fastest RAID 0 array to the finish line can save you a bundle. When rating RAID 0 arrays, speed weighed heavily but we also considered other factors such as price, features, and support. We've separated the PCI products from the NuBus ones to compare the products fairly. Here's how the field panned out.
Rating Product Price Support Features Speed
PCI Systems
(4.5 out of 5 mice) Average/Very Good mmmmh MicroNet DataDock Wide
Storage System 2 3 3 3
This array ($5,860 list, $5,275 estimated street price) is the fastest system, thanks in part to its dual-channel design. Not only does the system provide top speed but it also comes in a well-designed enclosure, and MicroNet provides customer support 24 hours a day, seven days a week.
(4 out of 5 mice) Average/Very Good mmmm DGR Ultrastar 4.0 3 3 2 2
(4 out of 5 mice) Average/Very Good mmmm MacProducts Magic
4 GB RAID 3 3 2 2
(4 out of 5 mice) Average/Very Good mmmm Microtech BLUE
Streak 3400 2 2 3 3
(3.5 out of 5 mice) Average/Very Good mmmh Conley Cobra 2 2 2 2
(3.5 out of 5 mice) Average/Very Good mmmh FWB SledgeHammer
4100 FMF-W 2 2 2 3
(3 out of 5 mice) Average/Very Good mmm CharisMac Anubis
4 GB RAID 2 2 2 2
(2.5 out of 5 mice) Average/Very Good mmh DPT SmartRAID 1 2 2 1
NUBUS and Built-in-SCSI Systems
(4 out of 5 mice) Average/Very Good mmmm Optima DisKovery
4200W 2 2 2 3
The Optima DisKovery 4200W ($4,155 list, $3,950 estimated street price) uses dual ATTO SiliconExpress IV cards to provide high speed. This overall-best NuBus performer comes with easy-to-use software that also supports RAID 5.
(3.5 out of 5 mice) Average/Very Good mmmh Conley Cobra 2 2 2 2
(3.5 out of 5 mice) Average/Very Good mmmh FWB SledgeHammer
4100 FMF-W 2 2 2 2
(3.5 out of 5 mice) Average/Very Good mmmh MaxConcept MaxRAID
SW4100e 3 2 2 2
(3.5 out of 5 mice) Average/Very Good mmmh MicroNet Raven Pro
NuBus Wide 2 3 2 2
(3.5 out of 5 mice) Average/Very Good mmmh Mirror RAID 2 2 2 2
(3.5 out of 5 mice) Average/Very Good mmmh Spin Whirlwind 2 2 2 2
(3 out of 5 mice) Average/Very Good mmm La Cie Joule RAID 2 3 3 1
(2.5 out of 5 mice) Average/Very Good mmh xiStor xi.RAID 2 2 2 1
3/ Outstanding 2/ Acceptable 1/ Poor
Listing Is Alphabetical Within Groups of Equal Mouse Ratings.
The Need for Speed / the Top Finishers and the Also-rans
Perform the same tests as MacUser Labs by using your own copy of MacBench 2.0, available online from ZD Net/Mac. See How to Reach Us for instructions on accessing ZD Net/Mac.
Desktop publishers, graphic artists, and video producers can never have too much speed when they're working with large files. Fortunately, a RAID 0 array can cut down on a lot of the thumb twiddling that normally happens while, say, you're opening large files. To compare their speed, we ran the arrays through a series of real-life tests. First, we timed how long it took to open a file in Adobe Photoshop 3.0.1 (Photoshop 3.0.4 was not available at testing time) in order to find out how well each array handled a large data flow. Next, we rotated an image in Photoshop -- a time-consuming, common, and disk-intensive task. Our third test, a Finder copy, also tested throughput and let us compare the different Power Mac bus architectures as well as the difference between the Finder that ships with the older Power Macs and the revised one that ships with the Power Mac 9500/132.
We also ran the MacBench 2.0 1,024K-sequential-write test. A high rate signifies an array that offers fast, continuous throughput, such as that needed for processing digital video.
All PCI RAID systems were tested with a Power Mac 9500/132, and all NuBus systems and the La Cie Joule RAID were tested with a Power Mac 8100/80. The arrays are listed here in order of overall speed.
Products Tested Photoshop Open Photoshop Rotate Finder Copy 1,024K Sequential Write
For this test, we timed how long it took to open a 32-MB file. Since Photoshop is less dependent on caching, this test indicates each array's raw throughput.
We rotated a 32-MB file 90 degrees counterclockwise. This common task is highly disk-intensive. On PCI Macs, Photoshop 3.0.4 should provide even faster results.
We performed a copy of a 32-MB file in the Finder. PCI Macs not only benefit from the hardware but also the new Finder is more efficient for transferring files.
We wrote continuous streams of data to the arrays. A high score indicates fast sustained throughput of large files, such as that needed for digital video.
Explaining RAID / What You Need to Know About the Technology
RAID 0
RAID 0 provides the fastest throughput, because it allows a file to be striped -- that is, written across two drives. RAID 0 provides the lowest cost per megabyte, because it offers no fault-tolerant features; all the tracks of the hard disks can be used for storing data. Be sure to implement some backup system -- if any drive in a RAID 0 array fails, all is lost.
RAID 1
Also known as mirroring, RAID 1 is the most fault-tolerant of the RAID levels, since a second drive contains a complete duplicate of the first. RAID 1 lets you get back to work instantaneously should a drive fail, since you don't have to restore anything. RAID 1 is expensive, because half of your storage is devoted to providing the mirror. Read times are shorter, because more than one drive reads the same file simultaneously, but write times are longer than with RAID 0, because all drives must be written to.
COST AND NOMENCLATURE aside, RAID provides advantages in two key -- and completely different -- areas: for networks, where file servers must constantly be up and running, and for graphics and video, which require ever greater speed for accessing and processing large files. RAID 0 through 5 are all about trade-offs: Each level provides a different balance of speed, reliability, and total cost per megabyte. Only RAID levels 0, 1, and 5 are prevalent on the Mac.
RAID 5
Best for networks, RAID 5 stripes data across multiple drives, a block at a time; parity information is calculated on a series of blocks and then written to a different drive. If one drive fails, then the file can be reconstructed from any other drive. With parity information as well as data striped across all drives, overall throughput is faster than with RAID 1 (but still slower than with RAID 0).
Removable Beasts / the Importance of Design
It's not a one-size-fits-all world. The systems we tested run the gamut from small, sleek desktop units to tall stacks of drives. Which one you choose depends on your long-term storage strategy. For example, if you think you may eventually implement RAID 1 or 5, hot-swappability -- the ability to remove drives without powering down your system -- is a feature you'll need. Having removable drives is a prerequisite for hot-swappability, although not all arrays with removable drives support it (see the "Array of Features" table for information on which systems are hot-swappable). Here are your choices when it comes to case design.
Removable Drives
Some arrays consist of a stack of removable drives. You can slip hard-drive modules into and out of docks that contain the SCSI circuitry and connectors. Unlike with standard removable hard drives, however, you can't easily access the data stored on this type of array from another Mac. The second Mac must not only have the correct dock for holding the drives but it must also have the RAID software installed and, for all systems except the La Cie Joule RAID, it must have the proper controller card installed. Some arrays with removable drives do, however, let you hot-swap drives.
Expandable But Not Removable
With some arrays, you can expand your storage space by stacking drives in vertical or horizontal configurations. The drives are held together by interlocking clips on the enclosures. These systems do not have removable or hot-swappable drives, but they do let you expand the number of drives in your array without taking up additional desk space. Microtech offers a removable version of the BLUE Streak 3400 for a few extra dollars (we tested the nonremovable version).
All-in-one Units
These no-nonsense small-to-medium-sized desktop boxes contain the drives that comprise the array in a single enclosure. If you don't plan to make any changes to your RAID configuration, these systems will work quite well. Note, however, that with these all-in-one units, you'll never be able to upgrade to RAID 1 or 5 with hot-swappability. Another drawback is that one drive may fail, but the other may still be good. Replacing the failed drive can be cumbersome. To add more capacity, you can daisy-chain additional drives through SCSI.
An Array of Features / Sorting Out the Specifications
There's more than just speed to consider when you're buying a RAID 0 array. The arrays we tested vary not only in speed but also in the features they offer. Some vendors, for example, use removable drives, whereas others use drives that were meant to find a permanent home on your desktop. There's also the issue of cards and channels -- do you want a single card, dual cards, or a single card with dual channels? How much capacity do you need, 4 or 8 GB? And the software package bundled with the array is important too -- some vendors license packages from third-party vendors, whereas others bundle proprietary software. Vendors also vary in the type of support they offer -- for example, toll-free tech support and/or five-year warranties.
Without a doubt, sorting out the prices, channels, software bundles, and other features can be harrowing. Our table can help you figure out exactly what each product offers. Check out "The Bottom Line" for our recommendations and ratings.
List price Estimated Capacity RAID software SCSI-2 interface RAID Single Dual 16-bit bus Removable Hot-swappable Toll-free Warranty
street price tested card Level 5 card cards per channel drives drives tech support
La Cie Joule RAID $3,999 (direct) NA 8 GB La Cie Joule RAID NA (uses Mac's NA NA + 2 2 5 years
built-in SCSI)
MaxConcept MaxRAID SW4100e $3,800 $3,540 4 GB Trillium Research Remus ATTO SiliconExpress IV 2 2 2 years
MicroNet Raven Pro NuBus Wide $5,370 $4,655 4 GB MicroNet Raven Pro Manager MicroNet Raven Pro 2[[lozenge]] 2 2 2 3 years
Mirror RAID $3,999 $2,700 4 GB Trillium Research Remus ATTO SiliconExpress IV 2 2 2 2 years
c Optima DisKovery 4200W $4,155 $3,950 4 GB Optima DiskArray RAID ATTO SilconExpress IV 2 2 2 1 year
Spin Whirlwind $3,599 (direct) NA 4 GB Trillium Research Remus ATTO SiliconExpress IV 2 2 2 5 years
xiStor xi.RAID $5,560 (direct) NA 8 GB ATTO ExpressStripe ATTO SiliconExpress IV 2 2 2 2 + 5 years
c MacUser BEST BUY
*32-bit bus per channel per card. +8-bit bus per channel. [[section]]$5,335 without the hot-swap feature. [[lozenge]]Dual channels on a single card. **Optional.
