Developing something new has the implicit assumption of the inability to achieve expectations. For startups, the risk is company survival. For IBM, on the other hand, the risk is simply that a particular investment did not pay off. But since IBM gets to employ the power of large numbers, if one project fails, that lost investment is more than recouped through those projects that succeed. And breakthroughs that push the technology envelope can also represent future product differentiation, illustrating why many larger vendors are more than willing to accept a measure of risk. (Soap box alert: That lesson should be applied to basic research on a societal basis, as well.)
One of the biggest questions about storage and one that should generate a lot of excitement is: What is the future of storage media? In other words, what technologies will eventually replace the tapes, hard drives and SSDs people really depend on today? IBM's work in this area is included in the Solid State Information Systems Initiative. This work is characterized under the label of storage-class memory (SCM), which is the company’s term for a new class of data storage and memory devices.
SCM collectively covers about 10 new technologies currently under development that will emerge in the market in a timeframe ranging from the near future (say, in two to three years) to further out (2020). SCM’s key characteristics are solid state (no moving parts, unlike disk), short access times (DRAM-like, within a factor of 10, low cost per bit (disk-like within a factor of 10), high write-endurance (which means that is it suitable for memory-like uses) and non-volatility.
There are several reasons why development of new SCM technology is mandatory if storage is to keep up with the performance and capacity requirements of future systems. The first is that the gap between the performance of disk (as measured in latency) and the rest of a large-scale, high-performance computing system is already at an ever-more-unacceptable five orders of magnitude (and recall that 105 is 100,000 times), and that gap is widening. In addition, IBM is already envisioning exascale systems that are capable of 1,018 operations per second. That requires that current obstacles in the form of energy consumption, space requirements, and the cost of the memory and storage hierarchy be overcome. The goal of IBM’s SCM technology efforts is to overcome those hurdles, and the company believes SCM has the potential to revolutionize data centers by 2020.
However, let’s not see breathless headlines that proclaim that tape (or disk or RAM or flash) is dead. All of those technologies likely have long lives ahead. Even when SCM becomes prominent on the storage stage, the older technologies will still have roles, although not the glamorous leading parts that they once took.
