Here's what storage vendors want you to believe about serial ATA (SATA) technology: It's a faster, more reliable...
version of the standard parallel ATA interface. It will eventually replace the older technology, and it's breathlessly awaited by the storage community. That's about two-thirds right.
"Serial ATA will be the standard interface in two years or less," says John Monroe, a vice president at Gartner Inc., Stamford, CT. "We'll see the death of parallel ATA."
But "breathlessly awaited by storage managers?" When was the last time you ever heard of someone becoming breathless over a computer part? (Perhaps one that didn't work.)
"We're sold on SCSI, and we're not really looking to swap what's reliable for what's slower and unproven. Call me crazy, but if it ain't broke ..." says Rick Bauer, chief information officer at The Hill School, a private preparatory school in Pottstown, PA.
Monroe says that this attitude isn't unusual in an enterprise storage setting. Enterprise storage markets are the ideal definition of conservatism, says Monroe. "Change comes slowly to the glass house."
Storage managers in those environments are loathe to buy a pig in a poke, says Ted Vojnovich, network-attached storage (NAS) systems designer and architect at IBM Storage Systems Group in Tucson, AZ. "I've been on several different customer councils, and one of the things we run into talking about serial ATA is that the IT guys want to see that it has proven reliability. They want to make sure that it does prove to be as reliable as Fibre Channel or SCSI," he says.
Most products aren't on the market yet, or are very newly introduced. Seagate has its Barracuda drive, and companies such as Intel and 3ware have serial ATA RAID controller cards. But in general, serial ATA has been slow to hit the market. "I initially forecasted at least 200,000 drives delivered this year, and there's been nowhere near that," says Gartner's Monroe.
Trickling to market
No matter how long it takes the technology to trickle into the market, it will arrive eventually. And as disk drive manufacturers hit high speed with their production volume, big subsystem makers such as Dell, Compaq, and IBM will have a steady supply of serial ATA products to add to their products. It's clear that serial ATA will eventually have a significant impact on the storage market. The questions are: Where will it make its inroads, and what will this mean to storage managers?
The general wisdom is that serial ATA will start to broadly ramp up over the next year as Intel comes out with a chipset supporting serial ATA, and drive production reaches critical mass. Jason Ziller, a technology initiatives manager at Intel Corp. who chairs the serial ATA working group responsible for developing technical specifications, says he expects to see it first on high-end desktop units and low-end servers. "Serial ATA will ramp top to bottom in desktops but bottom to top in servers," he says.
Price will determine how fast serial ATA creeps into IS shops. Most experts agree that when serial ATA prices rival parallel ATA, serial ATA will take off. After all, faster and more reliable technology at the same price - what's not to like?
"The whole disk drive market is driven by cents per megabyte and capacity," says Robert Cecil, Ph.D., the network director at the Cleveland Clinic Foundation, a hospital in Cleveland, OH. "So if SATA comes in as a cheaper solution, it wins hands down - end of discussion."
IBM's Vojnovich says that most corporations start small. He expects first deployments to be in a specialty application or a remote office "to see if it has legs." This translates into lower-end departmental-level NAS or server platforms, he says. "This would be the guy with a fair amount of disk, but not data center amounts."
|A sampling of serial ATA products|
Once serial ATA gains a toehold at the low end, the big question is whether its price/performance trade-offs will render it acceptable for high-end storage systems. "Arguments are intense as to how far up the storage food chain serial ATA will go," says Gartner's Monroe.
Most industry experts expect to see it first in low-end enterprise-class subsystems. For example, Zambeel Inc. will be rolling out a serial ATA version of its enterprise NAS system early next year, says Karl Schubert, the senior vice president of engineering at the Fremont, CA-based storage manufacturer. In fact, Schubert expects to see serial ATA in fairly high-level systems. "SCSI and Fibre Channel don't have to be the only solutions," he says. "If what you're doing requires a high OLTP rate, you'll use Fibre Channel. But the actual number of applications and quantity of disks for uses like that is relatively small, compared with the amount of data kept as file data and used for normal applications."
There, he says, serial ATA will do fine, particularly as the technology becomes seasoned. Schubert points out that similarly, parallel ATA was embraced by big companies such as Compaq and Dell as the technology became more reliable. "These guys went back to parallel ATA because of its price and value," he says.
SATA enables new applications
Serial ATA is a natural for disk-to-disk backup applications. Many experts expect to see serial ATA in what StorageTek calls the "mezzanine layer" of disk subsystems (i.e., above tape systems and below the very high-end storage systems).
Arun Taneja, a senior analyst at Enterprise Storage Group in Milford, MA, sees this storage niche booming in the near future. "We have a tsunami of new types of data that needs to be stored in different ways than transactional data," he says. New laws require Health Insurance Portability and Accountability Act of 1996 (HIPPA) and financial services data to be stored online, and recoverability needs to be in seconds, not days.
"A scenario like that begs for serial ATA," says Scott Robinson, the chief technology officer at Datalink, a storage outsourcer based in Minneapolis, MN. "Subsystem serial ATA for most applications will provide all the performance you need for less money. There's almost no reason to use a higher performance, higher priced subsystem if you have serial ATA," he says.
"We expect to have serial ATA at close to a penny a megabyte by middle of next year, says Robinson. Compare that with today's typical midrange RAID systems based on Fibre Channel. They're typically three to five cents per megabyte. If you back up terabytes of data, that price differential can be pretty dramatic," he says.
Robinson takes it one step further, saying that disk-to-disk serial ATA will soon be cheap enough to rival tape cartridges. "By 2005, he says, "the per-megabyte cost of ATA disks will be close to converging on the cost of tape cartridges."
Steve Luning, the director of storage architecture and technology at Dell's Enterprise Systems Group agrees, saying many vendors will start enhancing disk-to-disk backup features to be more competitive with today's disk-to-tape backup. "Clearly, putting tape in a box and putting it off-site isn't something you do with disk," he says. But if you want to replicate data for safety, it might make more sense to use a serial ATA farm of disks than a tape library. Of course, disks offer incremental backups and random access.
Luning pushes the possibilities of serial ATA even further, to what he calls disposable or self-healing storage. Says Luning: "Because serial ATA is so cheap, we can ship storage arrays with a few extra drives in them."
He envisions one scenario with a shelf of 15 drives. "You package 15, but only offer 13 for use. Any time a disk fails just switch over to one of the extras." This becomes feasible when technology is cheaper than service costs. Luning says that Dell is working on figuring out the price point at which this scenario will work.
Much has been made of the fact that serial ATA isn't backwards compatible with parallel ATA or SCSI. But is this really a big deal? Not for Datalink's Robinson: "It's not compatible from a hardware connectivity perspective, but it will be compatible for software drivers written for parallel if you get an adapter card, and those are coming out as we speak."
|Serial's selling points|
"With an adapter in front of those drives, I don't see a lot of difference," says Zambeel's Schubert. "The attractiveness of serial ATA is still greater than the pain of incompatibility, he says, adding that the cost/benefit ratio is clearly in the right direction."
Others, however, point out some possible complications. Gartner's Monroe, for one, is skeptical that the resolution of the software issue is as simple as an adapter card, particularly on direct-attached storage (DAS). He says that the drive manufacturers all maintain thousands of lines of propriety microcode for each family of drives. "For one family of SCSI drives, Seagate may maintain 20 different versions of microcode to satisfy the various requirements of OEMs," he says. DAS systems are more likely than fabric-attached storage (FAS) to use this microcode to fine tune the drive characteristics with the peculiarities of the system. "It's going to be very hard to plug-and-play replacements in that kind of environment," Monroe says. "On the other hand, if you're connecting to FAS, it's going to be pretty easy to use serial ATA instead of SCSI internally on the other side of the subsystem."
Vojnovich raises another issue when he suggests that migrating data from SCSI to serial ATA drives could also present problems. When you migrate data from legacy systems, the issue becomes the intelligence that's managing the disk, he says.
"As far as old files on parallel ATA, there are tools to help move them over: mirroring, backing up to tape and restoring it to a new unit." The two technologies' similarities make migration a relatively simple matter. "But if somebody wanted to take data on SCSI and read all the blocks onto serial ATA, that would be somewhat of a challenge," he says. "The formats and protocols would be difficult to reconcile."
Whatever the difficulties, it's clear serial ATA drives will have a significant presence in the storage markets in the near future. Monroe predicts the technology will hold 30% of the enterprise market by 2005.
"As we come out of this bad economy, there will be a lot of pent-up demand, but not much budget," says Monroe. Storage budgets will be scrutinized far more closely. "Instead of one level of sign off, you might have five and if you're not proposing the least expensive option that will solve the problem, you'll have to prove why," he says.