Hands-On Testing: 16 Fast, Reliable RAID Subsystems

If network server downtime has you singing the blues, the disk array subsystems tested here will keep you and your organization up and running

Michele Guy

Your organization's network file server dies. Day-to-day operations are paralyzed. What do you do? This scenario occurs more and more frequently in today's office environments. However, the trends in computer use (e.g., centralizing data and applications on file servers and downsizing from mainframes to PC-size servers) mean that more companies are no longer tolerating server downtime--they want a solution. We tested 16 fast and reliable disk array subsystems that deliver multigigabyte storage and ensure that the data on your file serv er is always available. The price for this kind of insurance starts at about $10,000.

The disk arrays we tested employ a data storage technology called RAID (redundant array of independent disks). RAID addresses three key aspects of disk storage: (1) capacity, (2) speed, and (3) reliability. A disk array connects multiple smaller-capacity drives into a device that can appear to an OS as a large, single logical drive. The overall speed is better on these drives than on a large single drive because the heads on the smaller-capacity drives travel a shorter distance to perform read/write operations, and multiple drives support multiple simultaneous read/writes. RAID controller hardware provides data redundancy to improve reliability, either with a second mirrored copy of the original data or through various parity schemes; this allows a RAID array to continue to operate if one drive fails. (Unlike most other components in a computer, fixed drives contain moving parts that make them more susceptible to fail ure).

RAID was originally defined as having five different levels. Each level addresses the issue of data redundancy in a different way. RAID level 1, which mirrors data, and RAID 3 and 5, which store parity information (also known as ECCs, or error-correction codes), are the most commonly used RAID implementations (for more on RAID level definitions, see the sidebar "On the Levels").

We configured the arrays in our test to use RAID 5, which gives you a reasonable trade-off between cost and performance. RAID 5 distributes data and ECCs across the entire array (see the sidebar "How Error Correction Works"). RAID 1 offers faster performance but at a higher per-megabyte price, because half of the total storage space is sacrificed to the mirrored data. On a typical five-drive RAID 5 array, parity information takes up only about 20 percent of total storage space. However, some performance is sacrificed because writes to disk must also include an additional operation to update parity information.

When RAID was first conceived at the University of California at Berkeley in 1987, the I in RAID stood for inexpensive . One of the original motivating forces for the RAID developers was to create the most storage for the lowest cost. They found it was cheaper to string several small-capacity drives together than it was to use a single, large expensive drive. Today, companies are more likely to use disk arrays for their redundancy features than to achieve cost savings. Large-capacity drives are no longer necessarily more expensive than an array made up of smaller-capacity drives. As the price-per-megabyte of disk storage continues to fall due to ever-cheaper drives, more users may find a RAID 1 mirrored drive configuration as economical as a RAID 3 or a RAID 5 solution. Another trend may make the focus on RAID levels less crucial. So-called adaptive RAID controllers that dynamically select the best RAID level, using whichever level is optimal for a given set of data, may soon be available.

---

HOW TO USE THIS GUIDE

We selected the best disk array subsystems by evaluating speed, features, and usability.

Overall Score: The Overall score combines a product's weighted scores for performance (i.e., speed), features, and usability. Performance counted for half of the overall score; features and usability each was one-fourth of the overall score.

Features: We evaluated the disk arrays on their features (e.g., warranty length and coverage), number of redundant and hot-swappable components, support for a hot spare drive, and alarm types.

Usability: Usability was judged on the quality of documentation, ease of configuration, and the ease with which the array was able to recover from a single drive failure.

Performance Index:

Overall: Relative overall speed on a scale of 1 to 10.

Single-Thread and Mu lti-Thread: Relative speed on a scale of 1 to 10 in a single-thread and a multi-thread environment.

---

A Pillar of Reliability

illustration_link (70 Kbytes)

REDUNDANT FANS
Self-contained units that cool the array. If one fails, the other will keep operating.

FRONT-PANEL DISPLAY AND KEYPAD
Depending on the manufacturer, a keypad with LEDs can give the current status of the array and let you configure the array and perform maintenance (e.g., a rebuild).

DRIVES IN INDIVIDUAL DRIVE SHUTTLES
We tested arrays with five half-height [3-1/2-inch form factor] SCSI drives of 2-GB capaci ty each. Arrays are designed to let you easily install and remove drives.

INDIVIDUAL DRIVE KEYLOCKS
Some models prevent unintended drive removal with a keylock for each drive; typically, the keylock must be in the locked position for the drive to operate.

ENCLOSURE KEYLOCK Depending on the design, the RAID cabinet keylock prevents entry either to just the drives or to the drives and other components [e.g., power supplies and fans].

SCSI BACKPLANE
Each drive connects to this when installed in the RAID enclosure.

INTERFACE CONNECTION TO THE HOST
On most of the units, this is a SCSI-2 Fast/Wide 68-pin female connection on the back of the RAID enclosure.

REDUNDANT POWER SUPPLIES
Self-contained units that supply power to the array. If one fails, the other will keep the array going.