Serial ATA: Coming Soon to a Data Center Near You?

Industry watchers say that solutions based on the parallel ATA or Serial ATA standards are coming on strong, especially in disk-to-disk backup.

Experts expect that Fibre Channel will almost certainly remain the dominant host connectivity standard for some time to come, but say that ATA-based solutions have their niche—yes, even in the data-center.

Market research firm International Data Corp. estimates that 554 petabytes (PB) of storage were shipped through the first three quarters of 2003, with another 200 PB estimated for Q4. Because ATA is the interface of choice for most of the hard drives sold in desktop computers, notebooks, appliances, and other devices (such as TiVo video recorders), it’s difficult to determine how much ATA capacity vendors are shipping to support enterprise servers.

According to Chuck Hollis, vice-president of storage platform marketing with EMC Corp., it’s a not-insubstantial amount. “By mid-year of 2003, we had done seven petabytes of ATA with Clariion [one of EMC’s product lines], which is 7,000 terabytes,” he comments. “We expect this [capacity total] will grow even more this year and next.”

EMC ships parallel ATA devices in its Clariion CX400 and CX600 systems, Hollis says, which have a nominal rotational speed of 5,400 RPM. The new SATA standard is much faster, debuting at 150 MB/s, and expected to reach disk-to-host transfer milestones of 300 MB/s and 600/MBs over the next two years, along with rotational speeds of 10,000 RPM.

EMC doesn’t currently ship SATA-equipped Clariion arrays. There’s a good reason for that, Hollis allows. “We look at ATA as a disk technology, so the interface isn’t very interesting to us,” he notes. “We have our choice: We can buy it as parallel ATA or Serial ATA, but which costs less? As Serial ATA volume builds, there may be a crossover point in the future [where it’s less expensive than parallel ATA], which is why we designed the Clariion to support both.”

According to Arun Taneja, founder and consulting analyst with storage consultancy Taneja Group, ATA-based solutions have initially found uptake in disk-to-disk backup and recovery, as well as in so-called “nearline” storage roles, where companies store data that would otherwise be written to tape on (relatively) inexpensive ATA disks for faster access and retrieval.

“We just finished a major end-user study on next-generation backup and restore, and I think the real answer is that some are using ATA [for disk-to-disk backups], although there hasn’t been a really massive deployment yet mostly because these things are relatively new. In fact, many of the drives still have parallel ATA,” Taneja comments. “I still expect that the first quarter and second quarter will continue to be evaluation quarters, but I expect that there’s going to be a serious line of lift-off starting in the second half of 2004.”

If vendor response is any indication, ATA-based solutions certainly have traction. IBM Corp. last October announced a new SATA-based addition to its TotalStorage product line, the FAStT EXP100 Storage Expansion Unit. At the time, Big Blue said the new SATA expansion was designed to provide up to 28 terabytes of capacity for its TotalStorage FAStT600 arrays and up to 56 terabytes of capacity for TotalStorage FAStT900 systems.

IBM’s FAStT100 expansion unit didn’t actually start shipping until December 19, 2003, but Big Blue has already had some success with it, says Craig Butler, brand manager for midrange brand marketing with IBM Storage Group. “We did sell a bunch of them at the end of the fourth quarter, including some into Unix accounts, mostly scientific computing research centers,” he confirms.

To the traditional bread-and-butter applications that are de rigueur for ATA-based storage—for example, backup and recovery and nearline storage—IBM has added a new wrinkle, targeting regulatory compliance efforts. “Our initial positioning is that these [SATA drives] are excellent for a retained data solution, the stuff for Sarbanes-Oxley, HIPAA, other legislation, so that you can get in line with regulations that specify how long you have to retain things like e-mail and internal documents,” Butler explains.

One reason for the popularity of ATA storage for specific applications is its cost: According to IBM’s Butler, the SATA drives Big Blue uses in its FAStT arrays cost less than half as much as Fibre Channel drives. Standard parallel ATA drives are even cheaper, and as economies of scale come into play and SATA production ramps up, this ratio could drop drastically.

At the same time, he concedes, there are other factors that tend to offset the low cost of the SATA hardware. “The SATA drawer [used in the FAStT array] is more expensive though, although this will be changing as volumes for SATA and SATA drawers go up across the industry,” he acknowledges, adding: “The combination of drawer plus SATA drives though, at the same total capacity point is still thirty to fifty percent lower than Fibre Channel depending on how you configure it. On the other hand, the Fibre Channel drive [gives] … more performance and can be used in more applications, such as transaction processing.”

Neither interface is deemed acceptable for OLTP and other write-intensive applications because of the slow rotational speed of most ATA and SATA drives (many vendors are shipping 5,400 RPM units, while IBM ships 7,200 RPM drives in its FAStT arrays). On the other hand, ATA and SATA drives boast excellent sequential read and write performance, which makes them ideal for backup and recovery applications in which data is written sequentially to, and recovered sequentially from, disk.

Another area in which the jury’s still out in the ATA-versus-SCSI or Fibre Channel debate is that of reliability. Typically, SCSI and Fibre Channel drives have boasted reliability ratings that are much greater than those of parallel ATA units. EMC’s Hollis says it’s a question of intelligently deploying ATA-based solutions to support the applications for which they’re best suited. “I think most people are commenting about the different duty cycles” of ATA vis-à-vis SCSI or Fibre Channel, he says. “Certainly, if you go and try to do a direct replacement for high performance server drives and plug in an ATA drive, they’re not designed for that kind of usage. But keep in mind what’s going on with the duty cycle in the backup and restore device, however: They wake up, they read or write lots of data sequentially, and then they power down. These duty cycles are very comfortable with what ATA does best.”

Besides, argues storage consultant Taneja, the low cost of ATA or SATA drives, coupled with the use of fault-tolerant RAID solutions, can help to nullify concerns about reliability. “Let’s say there’s 60 drive in this box, there’s 60 Fibre Channel drives or 60 SATA drives. I’ve got raid and all of that stuff built in already, so if a drive dies, because it’s (relatively speaking) less reliable—so instead of having two million hours of rated life it has only one million—it might fail a little more often, but heck, I’m still protected,” he argues.

Industry watchers seem to be torn over the future of high-end storage. Some, like Taneja, believe that the maturation of SATA—coupled with the plummeting price of SATA drives as production ramps up—create a kind of inevitability for SATA, especially in view of considerably more expensive SCSI and Fibre Channel solutions. “Serial ATA is going to take over the enterprise as well as the mid-tier stuff, like a storm. We just haven’t started to see it taking off yet, because the SATA stuff is just starting to ship,” he says.

Taneja says that the work of startups will do much to increase the performance, interoperability and—perhaps most importantly—reliability of SATA drives, much like they did with Fibre Channel.

Others, like IBM’s Butler, believe that a rival standard—called Serial Attached SCSI (SAS)—will displace both conventional SCSI and Fibre Channel to rule the enterprise roost. “By 2007, you might see a lot of Serial Attached SCSI drives and a lot of SATA drives, and maybe still some Fibre Channel,” he says, noting that Hewlett-Packard Co. is heavily committed to SAS. “I think SATA will expand rapidly to occupy niches like retained data and retained storage, and I think SAS will come in and take business away from SCSI and Fibre Channel.”

EMC’s Hollis rejects both prognoses, saying that good old Fibre Channel, which will soon scale to support 4 GB/s of throughput, will remain on top. “SATA in the back room? Performance and ATA don’t go together real well, and what we find in the enterprise environment is that when people want performance, they step right up to Fibre Channel.”

Most industry watchers agree that there’s a place for SATA solutions in the data center—when used judiciously. IBM’s Butler even suggests that the day could come when SATA solutions are used (again, intelligently) in tandem with mainframe systems. “Mainframes can connect up to retained data offerings, and I think they will over time. I think the economics are just too blatantly obvious. That’s why we’re working so hard on getting all of our software and disc options in a row,” he says, stressing, however, that “I don’t have anything, say, zSeries direct-attached to FAStT, today.”