SCALABLE ENTERPRISE RAID

By J.B. MilesSpecial to GCNSince the early days of computing, information systems managers have faced the double-edged challenge of providing fast access to mission-critical data while protecting it from loss or corruption. The challenge persists, even as high-profile technologies such as storage area networks and network-attached storage come on the scene.As a result, the market for external RAID subsystems, already healthy, will soar to nearly $25 billion by 2002, according to industry experts. A fair share of that market will include federal systems.RAID systems combine several, sometimes hundreds, of hard disk drives into one virtual storage system. A simple tower or rackmount RAID module may contain eight or nine disk drives, but also may be daisychained to other modules to form a huge system with hundreds of drives. In all cases, the host computer recognizes only one large drive through the RAID controller or controllers inside the enclosures.Think of a basic RAID system as an enclosure with eight or nine empty bays for holding disk drives. Add a power supply, cooling fans and high-speed SCSI or Fibre Channel interfaces for linking to host servers and other modular expansion units. Single or dual controllers usually come with each enclosure or expansion module to manage the drives it contains, making them appear as a single, logical drive to the host. The whole thing is run by RAID software provided by the manufacturer that tells the controllers how to behave.The products listed in this guide allow you to expand from a fairly small subsystem to enterprise-level RAID with relative ease. A small array can be built around a tower or rackmount configuration that holds between two and eight drives for a storage capacity ranging from under 100G to 250G or more, depending on the number and capacity of the drives that are used.By adding expansion modules to a unit, you can scale a system to 100 drives or more and create a boost in storage capacity into the 1T'1,000G'range.A RAID array can use tape or 5.25-inch magnetic disk drives. The ones listed here use 3.5-inch magnetic drives, which are the most common today. Some units with 3G drives are still around, but most current RAID arrays have 9G or 18G drives, which double the capacity without adding much to the overall space taken up in each drive bay.New drives with capacities of 36G and even 50G are coming on fast as well'though because the latter involve the use of more drive spindles, there is a slight performance price to pay in speed, especially in the 50G drive category. Thus, a modular RAID array with 100 18G disk drives can generate about 1.8T of storage capacity.Most of the arrays listed can be scaled up from a handful of gigabytes to one, two or more terabytes of storage. As such, they would be appropriate for a department that requires 250G to 500G of storage with the ability to expand the system, when necessary, to serve as an enterprise or data center array with even higher storage requirements.The extent to which you can scale a particular array depends on how many drives you place in your base module, the individual capacities of those drives and how many extension modules are allowed by the manufacturer.As for price, the more storage capacity and features you want'redundant components, dual failover controllers, fast Fiber Channel interfaces'the more you'll pay.Base models of highly scalable arrays such as Dot Hill Systems Corp.'s LR-5000/7000, Cambex Corp.'s Centurion 2000 series and Raidtec Corp.'s FibreArray series fall into the $10,000 range. But building them into high-end enterprise arrays with terabytes of storage capacity will cost $150,000 or more.Advanced arrays loaded with hundreds of drives and 3T or 4T of storage can cost $350,000 or more. Are they worth it? If your data is mission-critical, who can put a price tag on it? Many of the arrays listed in this guide provide full protection against data loss and high throughput speeds for a per-megabyte cost of 25 cents or less.RAID supplies two critical benefits: 100 percent protection against data loss and speedy input/output performance. RAID works by striping portions of a file across different drives. A single file may have one segment of data on Drive 1, more data on Drive 2, and so on. The data can be recovered via parity checking'parity is a mathematical formula for storing files on disks'and mirroring, which stores the same file on several drives.But in both cases, redundancy is key: If one drive fails, access to critical data isn't lost because it resides in various forms on other drives.Other protections RAID systems supply are hot-swapping and fault-tolerant components. Hot-swapping replaces failed drives or controllers without powering down the system, thus saving money and aggravation by eliminating costly downtime. Fault-tolerant RAID components include redundant power supplies, cooling systems, multiple host interfaces, dual buses and dual failover controllers, which direct the secondary controller to kick in immediately if the primary controller fails.Adding fault-tolerant components increases the cost of a RAID system, but they also provide virtually foolproof insurance against cataclysmic system failure and data loss.As for I/O performance boosts, RAID arrays alleviate the bottlenecks common in single-drive systems. Because RAID striping distributes segments of data files on disks throughout the array, each drive head can access the segments quickly and pass them through the system simultaneously. ''One of the best uses of RAID is for online transaction processing (OLTP), where many I/O requests compete for drive access and can bog down the server. But RAID storage is also used as a counterpart to enterprise SANs, high-speed, high-bandwidth storage networks that logically connect storage devices to ser-vers.When researching RAID systems you'll see the term RAID levels.Vendorspeak aside, RAID levels are descriptions of how multiple drives are connected and how they work together to protect data. There are five common levels: By itself, Level 0 amounts to almost RAID. RAID 0 distributes, or stripes, data across the array. It provides fast transfer rates, but the data isn't duplicated from disk to disk. So if one disk fails, the entire array goes down. Level 1 provides disk mirroring. A simple RAID 1 system consists of two drives on which duplicate information is written simultaneously. Level 1 provides dependable, albeit basic, data reliability because of its redundancy'two disks hold the same information. But it lacks parity checking, so corrupted data on one disk could corrupt data on another.. Known as striped mirrored array, Level 0+1 blends the speed of RAID 0 with the redundancy of RAID 1 for inexpensive data protection and fast throughput.. Level 3 combines data striping with a dedicated parity drive for error recovery.Three drives are required: one for parity and two for striping. This level also requires extra data overhead for parity checking, and when combined with other re-quirements makes it more expensive than RAID 0+1. . Level 5 stripes blocks of data and the parity checking formulas together across multiple drives, thus eliminating the need for a separate parity drive. This level is popular with users because it supports most high-end applications, including OLTP with its demanding read-write ratios. It is also dependable, but when a drive fails, recovery time is slow because the information on the failed drive must be reconstructed from the parity data on other drives.There are several other seldom-used RAID levels, including levels 4 and 6. Levels 10 and 53 amount to vendorspeak for RAID 0+1 and RAID 5 and 3, respectively.A white paper from PerifiTech Inc., a RAID array manufacturer in Hinkley, Ohio, suggests that when choosing a RAID level, consider the following factors:' The number of users and the amount of drive capacity needed' The importance of the applications and data to the cost of downtime and lost business' The size of the data blocks and whether they require direct or sequential access on the drives' The ratio of reads to writes to I/O activity, and the maximum transfer rate needed' The capabilities of the various RAID levels; a high number is not necessarily indicative of a better choice.XXXSPLITXXX-XXXSPLITXXX-