Take 32 of today's new 9Gb disk drives, pack them into a RAID device, and you get roughly 288Gb of storage. What do you do with all this storage?
Assuming you count every byte as usable storage, you could translate that into a 675-metre stack of Webster's New World Dictionaries. It could represent about 2,000 copies of Microsoft's Office 97 (OK, Office is a little chubby). And, it matches the capacity of about 57 IBM 3380 disk drives - the typical top-end mainframe disk drive of a decade ago. Oh, and each megabyte of RAID storage might cost about one-tenth what you would have paid for a megabyte of yesteryear's 3380 storage.
RAID is getting cheaper and gaining new capabilities, but it still isn't free. Often buyers must set priorities and balance functionality with price, and performance with capacity. To help buyers shop for devices that run with Unix or Windows NT servers, we asked experts for advice about prices, features and performance.
The key factor in determining prices will be the emergence of the 9Gb, 3.5in disk drive as the standard for RAID subsystems. List prices have dipped to as low as $1 per megabyte.
James Porter, president of Disk/Trend, says prices for complete RAID systems still haven't fallen as fast as basic disk prices because of the "overhead" costs related to complete system parts such as the frame, connectors and logic chips. But he says he expects even that to change after a coming market shakeout removes some smaller vendors. Then larger vendors will realise cost efficiencies from volume production, he says.
Analyst Andres Lofgren said: "I encourage users evaluating products to make sure they get competitive bids. But I also would caution them to not strictly look at dollar per megabyte. When you just look at dollar per megabyte, you aren't considering the performance of that box, its reliability or its redundancy. Price is just one aspect of what you have to look at."
He advises buyers to build forward pricing caps into their original contract to limit what they will pay in the future for items such as additional disks.
Many vendors promote extra functionality, such as remote mirroring, rather than price. And such functions aren't cheap. Some users, including those moving data from mainframe environments to Unix-based data warehouses, should expect to pay twice as much for full-function RAID as they would for basic RAID subsystems.
Expect more vendors to follow EMC's lead and deliver functions such as remote mirroring, which allows information systems managers to protect data from natural or man-made disasters by creating backup copies of data at off-site locations. People can also protect data by shipping tapes to an off-site vault, but the data written to disk since the last backup will have to be reconstructed from log files or could even be lost entirely. Remote mirroring enables much faster recovery after a disaster: seconds, minutes or a few hours at most rather than several days.
Gains in RAID performance will come partly from disk drive improvements, such as higher data rates, and partly from support of larger numbers of drives per controller, larger controller caches and faster interface connections. Disk drive seek and rotation speeds increase slowly from year to year, so applications that require small, random reads and writes won't benefit much from next-generation disk drives.
However, newer drives typically offer faster data transfer rates, which will improve performance on applications that read and write large sequential data blocks. Applications that require large block transfers could see benefits from expected vendor moves to high-speed Fibre Channel Arbitrated Loop (FC-AL) connections and the competing Serial Storage Architecture (SSA), or even the 40Mb/sec Ultra SCSI interface.
A key piece of advice for buyers when it comes to scalability is to look two years down the road. When you buy a system today, ask yourself, "What is my growth path for this? Will it keep up with demand for two years or whatever the life cycle of the system will be? Will it meet my needs, not only in terms of scalability but in terms of performance and reliability in the future?"
Fibre channel-based systems can help users deal with scalability issues in the future. One advantage of fibre channel is that its architecture could help managers string together modular arrays. That would make it easier to pilot-test an application using a minimum of storage and then add RAID boxes during a rollout.