In-Depth
Terabyte Drives Have Arrived
In less than 18 months, perpendicular recording has gone mainstream. It is one of those little-noted technology innovations, but one that is extraordinarily important.
A corporate communications operative from Seagate pinged me today to make sure I was ready for the introduction of Seagate’s terabyte-sized disk drives, something he said would happen during the first half of 2007. The new drives, he noted, will build on the same technology as Seagate’s 750 GB perpendicular magnetic recording (PMR) drive (available since April 2006) and which now enjoys an installed base of approximately 10 million units.
It struck me that, with very little fanfare and in less than 18 months, perpendicular recording has gone mainstream. It is one of those little-noted technology innovations, but one that is extraordinarily important.
It wasn’t long ago that the industry worried about the Super Paramagnetic Effect (SPE) and the hard stop it represented to improvements in areal density in conventional magnetic disk recording. The "SPE barrier" manifested itself in the practical issue of "random bit flipping" that occurred when bits were placed too close together (laid out end-to-end with their magnetic fields aligned longitudinally, or in "parallel" with the media surface). As the number of bits per square inch increased, their magnetic fields would collide with one another causing them to "flip" at random, corrupting the data they stored.
Without a lot of hoopla, perpendicular recording broke the SPE barrier and enabled the kinds of areal density improvements we see today. Hitachi, one innovator in PMR, wrote a "music video" about it, but the technology went mostly un-hyped in the media. (For anyone with the interest and the time, the music video can be viewed online at http://www.hitachigst.com/hdd/research/recording_head/pr/PerpendicularAnimation.html).
Perpendicular recording is essentially a technique, dating back to the mid 1980s, for writing data bits to disk so that their magnetic fields are arranged at right angles (hence, perpendicularly) to the media. This arrangement enables you to squeeze more data onto the same amount of available platter space than older parallel recording techniques, allowing an increase of the areal density (bit holding capacity per square inch) of a disk drive by many orders of magnitude.
The Seagate spokesman went on to note that the 1TB drive "will use fewer heads and discs than similar-capacity products we expect to see from our competitors. It is clear that fewer heads and discs, along with our proven perpendicular technology, can increase drive reliability, and also reduce operating temperatures, power consumption, noise, and weight." This is good from the perspective of IT managers worried about caps on their operating budgets, including electricity and HVAC: fewer, more efficient, disk drives holding greater amounts of data.
He added a few more bullet points to reinforce the key messages about the drives, emphasizing that they delivered "a best-of-breed combination of capacity, performance, and reliability." In my estimation, these points were part of a broader effort to reassure me about something that everybody seems to have been taking on faith—or simply disregarding: that PMR drives are just as dependable and trustworthy as older longitudinal devices. Seagate’s reasons for emphasizing dependability probably go back to the original perpendicular drive, in 2005, from Toshiba. Aimed at the mass consumer market, the 1.8 inch unit was problematic and notorious for its defects: a false start for PMR drives.
Big, Bigger, Biggest
While Hitachi made significant forward-looking announcements about PMR technologies and its own product line, Seagate was actually first to market. The company brought out its first crop of perpendicular drives, the 2.5 inch laptop hard drive called the Momentus 5400.3, rather stealthily. They used their own personnel as guinea pigs and used the drives in their own corporate laptops for many months to confirm their dependability. By January 2006, the company announced not only the general availability of the Momentus, but added that it intended to make PMR technology a part of most of its disk offerings by the end of the year.
In April 2006, Seagate began shipping the world's first 3.5 inch perpendicular recording hard drive, the Cheetah 15K.5, boasting a 300GB capacity. Operating at 15,000 RPM, the vendor said that the Cheetah PMR drive delivered 30 percent better performance than its predecessors with a data rate of 73-125 MB/s. At the same time, they announced the availability of Barracuda 7200.10, a series of 3.5 inch drives providing a slower rotational speed but increasing capacity to 750 GB.
At the Consumer Electronics Show in Las Vegas last week, Hitachi, which despite its catchy music video lagged behind the pack with the release of its first PMR drive (a 2.5 inch unit with a 160 GB capacity) in mid-2006, took the opportunity to show off a 1 TB drive that spokespersons said would become available in quantity later in the year. To my knowledge, Seagate did not show its next commercial drive at CES, but it did take the opportunity to articulate its plan to have available a Seagate 1 TB drive by mid-year. (Worth noting: Seagate announced in early December that it had decided to build its own substrate manufacturing plant in Malaysia for $282 million, ensuring sources of supply for disk media as demand continues to grow.)
It appears, therefore, that PMR drives are on the climb. Hitachi makes the most noise, but Seagate seems to be selling the most units and delivering capacity improvements at a faster clip than its competitors. Toshiba and Fujitsu have also introduced PMR technology drives.
The Big Impacts of Bigger Capacity
So, why should we care about this development? Drives are getting bigger and cost per GB is dropping annually: the SPE demon has been pushed back once again and the essential engine of disk economics continues unabated. So, what’s new?
Several things. First, assuming that PMR drives make their way into enterprise class arrays (they will, according to most vendors I’ve interviewed), then several opportunities and challenges are in the air.
As already noted, we will see an aggregate reduction in the size of enterprise disk arrays, certainly in the number of drive trays associated with any controller. Fewer physical disk drives might translate into simpler infrastructure configurations and fewer targets.
The reduced number of spinning rust devices will also reduce aggregated energy demands, a point that is increasingly fashionable even outside the "green data center" crowd. Because fewer drives will produce less heat, increased use of high capacity TB-sized drives will have the impact of reducing "environmental" costs for storage and lowering AC bills.
There are also challenges, of course. Hugely capacious drives represent a significant exposure to data loss. Put simply, lose a small capacity drive, lose a little data; lose a piece of media with a capacity of a TB or more, lose a lot of data. So, data protection becomes even more important than ever before.
As we already discovered with longitudinal SATA drives in the three or four hundred GB capacity range, RAID 5 is no longer a viable solution for protecting disks sporting high capacities. Rebuilding a 1TB SATA drive under RAID 5 will take a lot longer than you probably have to get your data back online in the wake of an outage.
Tape is keeping pace with large capacity drives—sort of. Linear Tape Open (LTO) generation 4 has just been approved for licensing and the latest Ultrium format specification doubles the storage capacity of the previous generation, increasing to 800 MB uncompressed/1.6TB compressed from 400GB/800GB in LTO 3. Transfer rates in LTO 4 are also improved, with speeds of up to 240MB per second in generation 4 promised by the spec writers—up from 160MB per second in generation 3 (assuming a 2:1 compression). In other words, with compression and under laboratory conditions, you should be able to back up a full 1TB drive in just over an hour. Restore will require three times that amount of time if the performance of previous LTO generations is any guide.
Another option that will certainly come into vogue is continuous data protection, probably accomplished through the virtualization of physical disk. This week’s announcement by DataCore Software of a range of virtualization-enabled solutions in this space provide an interesting bellwether of this trend: a TiVo for data that enables the ongoing backup of bits and near instantaneous restoration of a file—or a drive image—in the wake of a disk failure.
In addition to data protection, the issue of operational efficiency is yet to be addressed. How well will file searching and retrieval operations perform on huge PMR disks? Here, we are moving into uncharted waters.
Data access on current generation high capacity drives has thus far not been an issue. Comparisons of file-copy speeds on longitudinal drives versus PMR disk have demonstrated an advantage for PMR. However, performance differences pertaining to large file copies, file finds, and activities such as virus scanning have been miniscule between the two technologies. Could this be a harbinger of PMR performance deficits as drive capacities accelerate? Some say no: with data packed much more closely, the head will need to seek over less distance to find the data it needs. However, the same argument can be made that with more data locations to search on the media, finding files will be a lengthier process.
Hybrid on the Horizon
Time will tell which of these views will gain a foothold, but by then the issue may be moot. Last week, a group of vendors formed a new Hybrid Storage Alliance to develop standards around "hybrid drives" that combine magnetic disk with a buffer of non-volatile flash memory. The use of a capacious memory cache might help to reduce dramatically the actual number of seeks made to magnetic media because recently accessed data is maintained in the memory cache.
While initially aimed at improving the battery life of laptops running Microsoft’s new Vista operating system, the approach may become mainstream as disks grow bigger. At an International Disk Drive Equipment and Materials Association (IDEMA) seminar held in Santa Clara in early December, engineers were looking at the issue of growing drives to multi-TB capacities using PMR.
From a cursory survey of the presentations and papers at the seminar, it is clear that current generation PMR drives will need to implement significant design changes to support multi-TB capacities. Current read-write head technologies will need to change to better align read heads over increasingly small bit locations and to read very small magnetic signals against increasingly significant background noise. Also discussed by the engineers was the potential for using more memory buffers, especially as the dynamics of bit storage on silicon and magnetics begin to converge.
The Superparamagnetic Effect has not gone away with the arrival of perpendicular recording; it has only been pushed back. For this reason, Seagate and the other PMR drive makers have been right not to hype their new technology. Drives do not operate in a vacuum and the latest generation of terabyte drives is no exception.
Going forward, storage infrastructure designers will need to think about the devices they deploy with a "generational" view—matching drive technologies, chip technologies, and network technologies to workload requirements and budgetary realities to maximize benefits of greater capacity while minimizing risks. Achieving a balance is likely to require diligent effort and the consideration of an increasingly complex number of technical variables.
Your comments are welcome: jtoigo@toigopartners.com