RAID subsystems

They say if you sit long enough at a cafe on the Champs Elys'es in Paris,
everyone you know will pass by. I say, if you listen long enough to an information systems
manager, you’ll eventually hear the same lament—data by the ton and every single
user wanting immediate access to all of it at all times.

There is a solution to the problem and each systems manager knows it by name: a
redundant array of independent disks to call his or her own.

Basic RAID has been around for a decade or more, but it’s gone through a dozen or
so development cycles in that time, making it difficult even for experts to keep up with
the changes.

One thing is for sure: RAID is technology-intensive and usually expensive. Those two
factors make it critical to select the right one the first time out—a tough order for
systems administrators who lack the time to wade through pages of product literature about
competing RAID levels, interface options, redundancy requirements and the best software
for the arrays they have in mind.

Contrary to popular opinion, the principal purpose of a RAID is not security but data
accessibility. Every RAID system is a set of two or more independent disk drives
configured by the RAID controller to appear to the host computer as one large-capacity
disk drive. Although the host perceives the system as a single drive, the data is
distributed across multiple drives.

The arrangement provides fault-tolerant storage and alleviates the input-output
bottleneck of typical single-drive systems. The payoff to users is 100 percent data
availability, regardless of disk failure, plus very fast data retrieval.

Is RAID worth it? The average starting price of the systems in this guide is $25,000
and up, depending on configuration. A compensating factor is RAID’s low cost per
megabyte of storage, which is usually measured in pennies.

Products such as MTI Technology Corp.’s Gladiator 6300 and Procom Technology
Inc.’s Reliant 1000 Series may cost well over $50,000 when fully configured, but
their per-megabyte cost for storage can average less than 30 cents. Not bad, considering
the peace of mind that comes with RAID.

All RAID systems with two or more drives are scalable, but some are more scalable than
others. It is on high-end, scalable RAID systems for large workgroups, departments and
enterprises or data centers that this Buyers Guide focuses. Such products hold at least
two and as many as eight or nine drives in a single tower.

The most common drive sizes are 4.3G and 9.1G in the 3 1/2-inch form factor, although
18.2G 3 1/2-inch and 23G 5 1/2-inch drives are coming on fast. Depending on drive size, a
single-tower RAID system using 9.1G drives can scale up from 18G to 82G of capacity.

Rack-mount systems allow installation of more disk drives than towers do. Many systems,
such as Artecon Inc.’s LynxArray 5000, Clariion’s Series 3000 and Compaq
Computer Corp.’s StorageWorks RAID Array 7000, allow linking of tower or rack-mount
designs in a single architecture that appears as a single disk to the host server. Such
systems provide scalable storage of 18.2G to more than 600G, in many cases.

Other systems are scalable to an even higher order. LSI Logic Inc.’s MetaStor
system has 50G to 1.8T of storage capability. Western Scientific Inc.’s Cyclone RAID
Ultra and Winchester Systems Inc.’s FlashDisk RAID arrays can each be scaled up to
manage more than 1T of storage.

All arrays listed can meet storage requirements of typical workgroups—10G to 200G.
Many can be scaled to meet large departmental—100G to 500G—or even enterprise
and data center—500G to 1T-plus—storage requirements.

How much should you pay? The average starting price for these scalable storage
workhorses hovers around $25,000, but variables make firm prices hard to come by.

Price depends on the number, types and sizes of drives you require, the number and type
of chassis, such as tower or rack-mount, types of interfaces and number of controllers.
Most prices listed are for the minimum configuration available from each vendor. The
bottom line is, what you’ll pay for a scalable RAID system depends on how you load
it.

The main reason to use a RAID system is to improve data availability in case of a
massive disk failure. Purely electronic computer elements such as CPUs, device controllers
and network interface cards fail infrequently, if ever. However, mass storage systems such
as disk arrays are electro-mechanical devices with a much higher chance of failure. The
types, or levels, of RAID are nothing more than descriptions of different methods to make
data available even if one or more disks on the array fails.

The RAID Advisory Board has standardized several RAID levels that determine how
multiple drives are connected and how they work together to protect your data. Different
RAID levels support different tasks; no superiority of one level over another is implied.

RAID 0. At this level, data is split up and striped, distributed
evenly, across multiple drives. RAID 0 has very fast data read and write rates, and
because no data overhead in the form of parity checking is involved, it provides maximum
data storage capability.

On the downside, no data redundancy is provided—if even one disk fails, all data
is lost.

RAID 1. Known as disk mirroring, true redundancy, or avoidance of
single points of failure, is provided at this level. A copy of each disk is stored on a
separate disk. Data reliability is high because selective multiple drive failures can be
overcome. But if corrupt data is written to the original disk, it reappears on others.
RAID 1 is dependable and suitable for most storage applications.

RAID 0+1. Known as striped mirrored array, this level combines the
redundancy of Level 0 with the speed and high storage capacity of Level 1.

RAID 3. At this level, all data bytes are striped across all drives,
with parity blocks stored on a separate, dedicated drive. The parity bits provide for
error checking and allow reconstruction in case information has been damaged. Because it
involves extra data overhead and the use of an extra parity drive, RAID 3 is more costly
than Level 0+1, but it’s very dependable and useful for large file transfers.

RAID 4. Not as common as RAID 3, this level also uses a separate drive
to store parity blocks. It offers very high data read rates, but low write rates, making
it useful only for applications where many writes aren’t required.

RAID 5. At this level, blocks of data and related parity blocks are
striped across multiple drives. The write performance limitations of Level 4 are almost
eliminated and use of many drives is enabled, making this the most requested of RAID
levels. RAID 5 supports virtually all types of applications, including transaction
processing demanding high read-write ratios.

RAID 6. Also called dual parity, this level of RAID stripes data
across multiple drives. At least two levels of parity are striped along with the data or
stored on separate drives. RAID 6 provides the highest possible data reliability and fast
read and write rates but is costly to implement.

RAID 10. Vendorspeak for RAID 0+1.

RAID 53 . Ditto for RAID Levels 5 plus 3.

Implementing RAID makes potentially lost data available if a drive fails, but
doesn’t protect against potentially fatal system failures such as faulty power
supplies, overheated components or massive board failures. For protection against these
and other problems, some level of redundancy should be built into your RAID array,
especially if it’s serving a mission-critical environment. An MTI Technology white
paper suggests the following checklist:

You can avoid the vagaries and expense of field technicians if your RAID comes with
warm swap or hot swap capabilities. Older RAID systems had cold swap components—the
entire array had to be shut down and powered off before a drive or other component could
be replaced.

Warm swap designs require halting all activity involving the failed component, but the
power needn’t be shut down. Hot swap is the best of all the options. A hot swap RAID
design lets you remove and change a component while the system is still running.
Electrical glitches leading to system hang-ups and data corruption are avoided.

Hot spare drives sound the same as hot swaps, but are not. If any drive on a RAID
system fails, the information it contains must be rebuilt quickly on the replacement
drive, usually via the parity systems built into RAID Levels 0+1, 3 and 5. The system
isn’t providing full data protection until this happens, and it can take precious
hours for it to be fully implemented.

A RAID supporting hot spare or hot standby drives has one or more drives installed and
ready to go in case an original drive fails. The data rebuilding begins immediately,
reducing the amount of lost time.

With one exception, all listed RAID systems use Fast Wide SCSI-2, Ultra SCSI-3 or Fibre
Channel connection technologies. IBM Storage System’s 7133 Serial Disk System uses
IBM’s SSA connection design. Fast Wide SCSI-2 offers 16-bit data paths and a
20-megabyte/sec transfer rate, and Ultra SCSI-3 offers a 16-bit data path and a
40-megabyte/sec transfer rate.

Fibre Channel is a high-speed 1-gigabyte/sec duplex transmission medium that offers
100-megabytes/sec throughput when used for serial transmissions.

It denotes both the optical backbones and copper-based twisted-pair and coaxial
connections used in today’s advanced network designs.

Fibre Channel offers great flexibility for advanced RAID designs because it can be used
both by networks and mass storage systems such as RAID. Nearly half the listed systems use
Fibre Channel interfaces, at least as high-speed connections from the arrays to the host
servers.

The number is up from only a handful less than a year ago, and the upward trend is
likely to continue.  

J.B. Miles writes about communications and computers from Carlsbad, Calif.