Solid State Disk Will Go Mainstream in 3, 2, 1...
Solid-state disk, once considered a niche technology for ruggedized, industrial and military applications, is on its way to the mainstream. This is partly because of SSD benefits, which include performance, power efficiency, ruggedness and a lightweight, compact size. But other developments have also come into play, including technology and market developments that have begun to help this technology overcome its pitfalls -- namely capacity, reliability and price.
Because SSD is based on NAND flash memory chip technology, it has no moving parts, which makes it faster and less prone to mechanical failure than hard disk drives. Today, costs are shrinking faster than ever, thanks to market growth, new technology developments and vendors working overtime to accelerate their SSD development.
For instance, Intel Corp. has been working since early this year to turn around its troubled flash memory business, which has been hit hard by decreased consumer spending on MP3s, phones and other devices that use flash memory. To combat those losses, it announced plans to accelerate the introduction of higher-density NAND flash chips for enterprise storage arrays.
So, the question is not whether but when SSD will go mainstream. There are two main application areas where enthusiasm is growing about SSD: notebook computers and enterprise storage. According to IDC, SSD unit shipments will increase at 76% from 2007 to 2012, driven by the notebook, ultrasmall notebook and enterprise storage spaces. Revenues will grow at a 70% compound annual growth rate in that time period, to over $5 billion in 2012. Meanwhile, per-gigabyte prices in the SSD notebook market will decline 44% from 2007 to 2012.
The Notebook Wars
Currently, SSDs from the likes of Toshiba, Samsung and Micron can be found in notebooks from vendors such as IBM/Lenovo, Apple, HP, Dell and Toshiba. According to Krishna Chander, an analyst at iSuppli Corp., 2008 is an introductory phase, a time for resellers, consumers and suppliers to watch what happens when SSD rubber hits the road. He sees SSD notebooks becoming mainstream in the 2011-2012 time frame. By 2011, he says, SSD will achieve 27% penetration in notebooks, increasing to 35% in 2012.
The two drivers for the SSD boom in notebooks are capacity and price. Currently, average capacities in available notebooks hover at 64GB, but the two largest manufacturers of NAND flash memory are already moving to larger capacities. Toshiba began shipping a $2,999 notebook based on its own 128GB SSD in June and Samsung has promised to ship 128GB SSDs to hardware makers by the third quarter. Samsung has also said it will release a 256GB SSD in 2009.
Meanwhile, Intel has announced a 32-gigabit flash chip, produced via a joint venture with Micron. It will ship in the second half of 2008, and Intel plans to quickly move the chip into solid-state drives in the 80GB to 160GB range, according to Jon Stokes, a senior editor of the Ars Technica Web site.
At the current capacities, mainstream users will be slow to adopt SSD devices this year, Chander says. Things will change when notebooks transition past the 100GB level, though, and by 2011, Chander sees average capacities at 200GB, increasing to 350GB in 2012.
However, for true road warriors who value lightweights and extended battery life, even 32GB capacities are just fine. "Power savings is the key attraction," Chander says, "in addition to the immediate log-on and extended battery life." Since SSDs generally lack moving parts, their response to a computer log-on is almost instantaneous, meaning they go into "read" mode very quickly, Chander explains. Hard disks, on the other hand, have to achieve stable rotation-per-minute speeds before starting to fetch required data. This entails a slight delay in loading the operating system into the processor and memory from the hard disk, he says.
In addition, SSD extends batteries by 15 to 30 minutes, he says, and the drives use less than 1W, vs. 2W for a notebook hard drive, he says. They also resist what's called "drop shock" -- because SSD is not a mechanical technology, the drives can sustain impacts without breaking.
Zsolt Kerekes, publisher of online storage publication StorageSearch.com, believes that what he calls the "featherweight notebook" market will be a key one for SSDs, especially since SSDs can be more effective at accelerating application performance in notebooks than sticking with hard drives and boosting the speed of the CPU, he says. "The SSD provides desktop application performance in a low-weight, long-battery-life form factor that is impossible to achieve using microprocessor technology, where high speed means high power, fans, etc.," according to an article he wrote on the Web site. "In a well-designed notebook, the designer can use SSD acceleration to give the same overall application performance as would be needed by a traditional notebook design using hard drives and a processor with three to five times higher clock speed."
IDC analyst Jeff Janukowicz agrees that the ripest market in the near term for SSD is in sub-four-pound, sub-$1,000 notebooks like the HP Mini-Note PC (with a 4GB SSD) and the Asus Eee (with up to 40GB SSD). Consumer adoption of the lightweight notebooks will begin later this year, he says.
It Comes at a Price
The price differential is still a big factor for traditional SSD notebook buyers, however, and Janukowicz says it won't be until 2010 that SSD will become more mainstream for higher-end notebooks. Currently, SSDs cost about $3.45 per gigabyte, compared with about $0.38 per gigabyte for hard drives, and they add a minimum of $500 to a notebook's price tag. "When you look at a 160GB notebook that's $400 to $500 cheaper with a hard drive, that's what buyers will think they need rather than a 32GB or 64GB SSD," Chander says.
Chander expects SSD prices to reach 72 cents per gigabyte by 2011, decreasing to 31 cents in 2012. At the same time, of course, hard drive prices will also steadily decrease, and capacities will climb. Still, somewhere in the 2012 to 2015 range, he sees SSD prices dropping low enough that the benefits outweigh the price differential.
Additionally, because hard drives are mechanical devices and not semiconductor-based, they will always have a fixed cost of $40 to $50. So by 2016, Chander sees the two technologies reaching price parity.
A key way SSD manufacturers such as Intel, Toshiba, Samsung and SanDisk are lowering the price is by moving from single-level cell (SLC) to multilevel cell (MLC) technology. While SLC-based SSDs store one bit of data per memory cell, MLC drives store two, four or more bits per cell. Because this increases capacity per square inch, MLC can offer twice the capacity as SLC at the same cost.
However, it also lowers performance -- SLCs are roughly twice as fast as MLCs, according to Joe Unsworth, an analyst at Gartner Inc. Toshiba's notebook, which uses MLC technology, has write speeds of 40MB/sec. and read speeds of 100MB/sec., although Samsung is claiming higher speeds -- 70MB/sec. write and 100MB/sec. read on its 128GB SSD, also MLC-based. Still, hard drives typically read data at about half this speed.
MLC also reduces SSD reliability by an order of magnitude because the chip wears out faster. To compensate, vendors like Toshiba and Samsung have developed what they call "wear leveling" algorithms, which reside on the controller chip inside the SSD along with the NAND flash chip. Using the algorithm, the controller spreads writes across the drive to avoid wearing out any one portion too quickly. "We're in the first phase of MLC this year, and as time goes by, consumers will see how reliable it is," Chander says.
Manufacturers are also shrinking the size of SSD devices to lower costs and increase capacities. For instance, Intel and Micron's 32GB chip is based on 34-nanometer technology, which increases the density of SSD chips on the wafer -- which in turn lowers cost, Chander explains. According to Intel, the product will enable more cost-effective SSDs, instantly doubling the current storage volume of these devices and driving capacities to beyond 256GB in today's standard, smaller 1.8-in. size. But there is a trade-off, Chander says, because even as you squeeze out costs through miniaturization, the design becomes more complex.
A key development that will increase SSD performance is the release of Open NAND Flash Interface (ONFI) 2.0-compliant parts, according to Stokes. This type of flash has dramatically faster read and write speeds, he says, which will change the performance equation for everything from SSDs in the enterprise and in mobile devices, to consumer products like Apple's iPhone. Intel's new SSDs will be based on the ONFI 2.0 spec, so their transfer rate will be blazing fast, Stokes says.
The spec, released earlier this year, defines a high-speed NAND interface that delivers up to 133MB/sec. compared to the original NAND interface's 50MB/sec., which significantly hampers performance in applications such as SSD.
ONFI 2.0 reduces the time required to transfer data to and from the data buffer by using two techniques, according to the ONFi Working Group. The first is DDR (Double Data Rate) signaling, which transfers data on both the rising and falling edges of the clock signal, doubling the data transmission rate, a technique commonly used in DRAMs. The second is source-synchronous clocks that accurately latch signals, enabling higher frequencies to be realized.
Outside the R&D Lab
Of course, it's not all about technology. Behaviors and logistics will also have to change for adoption to increase, both among resellers -- which have to switch their products over from traditional hard drives to SSDs -- and end users, who need to grow into confidence in SSD, Chander adds.
Another factor is the performance variation among SSD chips. A series of tests from the Transaction Processing Performance Council showed that write transfer rates among SSDs from a variety of vendors ranged from 26MB/sec. to 113MB/sec. and read rates varied from 49MB/sec. to 142MB/sec. "Most of the SSD notebook implementations have not been very good, and muddling the issue have been concurrent bad experiences in the hard-disk notebook market due to Microsoft Vista," Kerekes says. "It makes many users wonder why they bothered to buy a new notebook at all."
The economy will also affect SSD uptake. While helping to push prices down, the current downturn will also likely produce both winners and losers in the featherweight notebook market. Assuming that the downturn is akin to the recession of early 2000, Kerekes sees the losers being products that offer mediocre performance, meaning not much higher than a notebook hard drive. "As notebook SSD prices drop, many of the 25 or so manufacturers making such products will try to gain market share or recoup investments before exiting the market," he writes on his blog.
The winners, meanwhile, will be products that are significantly faster than hard drives "because there is always a market for faster products, even in a recession," he writes on his blog. At the same time, the recession will put a damper on new notebook sales, making it unlikely that some of the high-penetration rates predicted by IDC will be achieved.
Meanwhile, Back in the Data Center
Flash memory-based SSDs are also making their way into the enterprise data center, where the performance and power savings of SSD are attractive. The main issues SSD needs to overcome in the data center are reliability, price and user acceptance. "It's still an untested market," Chander says. "IT managers have a lot of mission-critical data that they're not ready to store in SSD."
In EMC Corp.'s case, SSD is positioned as a high-performance "tier zero" in the company's Symmetrix DMX-4 array. "EMC is saying you can put frequently accessed pieces of your data set on SSD, to increase system performance," Janukowicz says. Less accessed data can remain on Fibre Channel drives.
EMC is using SLC NAND chips from STEC Inc., but it says it's also adding a lot of its own intellectual property to ensure reliability that's on par with Fibre Channel, as well as the ability of the array to take advantage of SSD speeds and I/O capability. The company expects SSD pricing to be on par with the very highest end Fibre Channel drives by the end of 2010.
Meanwhile, Sun Microsystems Inc. has unveiled a plan to offer 2.5-in. and 3.5-in. SSDs across its servers and storage products in the second half of this year. It also tweaked its ZFS operating system to handle SSD. In Sun's case, the role of flash memory is as a high-speed cache, Stokes says. Sun claims that it can dramatically boost performance while lowering power consumption by coupling an SSD-based disk cache with a pool of slower, cooler-running drives, he says.
The company believes there will be rapid adoption of SSDs for performance-intensive workloads in the next six to nine months, which Stokes sees as a compressed timetable but similar to the track Intel says it's on. Janukowicz agrees with this time frame. "We're seeing uptake even today," he says, "and we suspect that will continue into 2009 and 2010 as systems get more broadly accepted."
But according to Chander, the world isn't quite ready for SSD at the enterprise level. "I think it's aggressive," he says. "Vendors can offer all that, but they need to take into account the box vendors and IT managers who are buying it and whether they're comfortable displacing hard disk with SSD."
Give it two and a half years, he says. SSD vendors say their drives have a guaranteed life of five years, and once people cross the half-life of that and the system hasn't failed, "word of mouth will help rally the market, and SSDs will take off," he says.
Where SSD really shines is in read performance, Janukowicz says, although vendors are currently working to increase write speeds. "A lot of the development work is to improve write performance, as well as read performance, he says. According to Kerekes, the asymmetry of read-to-write IOPs will improve to 5-to-1 by 2010, but won't progress beyond 4-to-1.
Energy savings may also help close the price gap, especially in data centers where companies have overallocated storage to get a certain I/O performance. "You can replace 10 hard drives with one SSD and the same I/O performance," Janukowicz says. EMC says a 1TB array of solid-state drives would draw 38% less power compared to hard drives.
Also accelerating SSD in the enterprise is the drop in NAND prices, according to Stokes. "The combination of dropping NAND prices and increased capital expenditures on NAND production (up by 20%, according to iSuppli) will accelerate enterprise adoption of SSDs," he writes on the Ars Technica site. "Rising data center power budgets and networked storage latencies have already made the case for SSDs in the data center, but the SSD price drop and the ONFI 2.0 spec will serve to seal the deal this year."
In any case, according to Robin Harris, an analyst at the Data Mobility Group who also runs the StorageMojo blog, SSD at the enterprise level doesn't need absolute price parity to win against high-end Fibre Channel drives. "Getting within 30% should do it for most people," he says on his blog. "Their performance advantages are worth at least that."
Kerekes agrees that SSD should be seen as a disruptive technology; in other words, rather than looking at price per byte, users will eventually consider the technology's other value propositions. It's similar, he says, to comparing a bike to a car or the PC to a typewriter.
"SSDs are seeping into many market applications long before the time that would be predicted by analysts who cling to the outmoded model of cost per byte parity with hard disks," he says on the StorageSearch site. "The SSD market in the long term will be bigger in revenue than the HDD market is today."