Laptop manufacturers are using low-capacity SSDs to build basic, yet light and cheap ultra-mobile PCs, such as the ASUS EeePC. But now SSDs are appearing in enterprise computing applications as well. EMC announced earlier this year that it would add the technology first to its high-end Symmetrix storage systems and then to its Clariion midrange arrays. At the time, EMC argued that the move was worthwhile because while enterprise-class SSDs cost around 30 times as much per gigabyte as Fibre Channel hard disk, it also estimated that SSDs would perform 30 to 50 times faster.
However, hard drives built out of memory chips have been around for at least three decades. In fact, the Symmetrix announcement wasn't even EMC's first foray into SSDs. The company briefly sold solid-state storage systems for minicomputers during the late 1980s.
In the years since, other vendors, such as Texas Memory Systems, have built nice businesses out of selling super-fast drives for storing database tables and the like. For example, German insurance company DAK uses Texas Memory SSDs to eliminate the bottlenecks that occurred when 14,000 users all tried to log into its file servers at much the same time.
However, like the original SSDs of 30 years ago, those are very expensive units built out of dynamic RAM (DRAM), and they're hard to justify outside the most performance-hungry of applications. As a result, their adoption in Europe has been slow.
In contrast, the SSDs being adopted by EMC, ASUS and others are based on Flash memory, just like USB sticks and memory cards. Flash SSDs still cost up to 10 times more per gigabyte than cheap mechanical hard disks, but they're less expensive than DRAM. Plus, unlike DRAM, Flash is non-volatile, so it doesn't need backup batteries and hard disks to save its data if the power fails.
The other big change is that today we're all familiar with the concept of solid-state storage, suggests Juergen Arnold, the chairman of SNIA Europe. "It's an awareness game, and awareness has dramatically increased in the last six to nine months. Everything is driven by price, and when that matches the benefit gained," he says. "This year, SSDs have caught the attention of end users -- I think 08/09 will be when it goes mainstream. It didn't take off before simply because of price, but now we can see a trend when it becomes cheaper than disk."
As an example of how awareness has grown, he notes that this year's SNW Europe conference in October will feature some 28 presentations on SSDs, up from zero last year.
Flash has also overcome its perceived technical shortcomings. The latest chips are far faster than those of just a few years ago, and each block can stand many more write cycles before it wears out. In addition, modern Flash SSD controllers include sophisticated wear-leveling schemes to spread the load and dramatically reduce the likelihood of bit errors.
Industry analyst Hamish Macarthur, founder of Macarthur Stroud International, adds that the growing emphasis on green has played a part too. Flash is high-density storage so it takes less floor space, doesn't generate noise or vibration (though its cooling fans might) because it's non-mechanical, and consumes much less power than equivalent high-speed disks -- as long as the rest of the system is optimised to work with Flash.
"It's not just less operational power -- the heat distribution could be more attractive," Macarthur adds. "You still have cost as a fundamental issue, but I see it coming in because there's less moving parts and a perception of reliability."
Welcome to Tier 0
For now, SSDs are promoted as a Tier 0 of storage -- somewhere to store the working tables and all the other data that an application needs to access most frequently. This is work that might otherwise be done by 10,000 or 15,000 RPM Fibre Channel, SCSI or SAS drives, so in a sense it is SSD replacing spinning storage.
Arnold points out that in the past, disk performance was good enough for most users, and those niche customers who really needed performance could always buy DRAM-based SSDs. But performance demands are continuing to climb, and as they go past what can be met by 15,000 RPM disks, more and more systems will need that Flash-based Tier 0.
There's still a lot of uncertainty over where Flash fits, though. While EMC and others want it in the storage subsystem, vendors such as HP argue that modern arrays already have cache memory, which is similar in performance to SSD. They say SSDs would be better off close to the server -- built onto a blade, for example.
Arnold says adds that it is also important to recognise that although it looks like a direct alternative to hard disk -- which is how it is used in laptops -- the properties of Flash-based SSD differ from those of hard disk, just as hard disk differs from its predecessor, tape.
For example, Flash has no mechanical latency so it performs much better on random reads than hard drives do. However, there is still a limit on how many times a Flash block can be rewritten, and current Flash technology is around 10 times slower to write than it is to read, so it is best suited to uses that are read-only, or at least "write-infrequently, read-mostly."
"Some vendors are honest that it's not an optimal fit everywhere," Arnold says. "It is important to identify your usage case -- it is not a replacement for disk. In 99% of cases you will still use disk, but in 1% you can use it to accelerate."