As an industry analyst, I go to plenty of technology events. Some are big, aimed at a specific vendor’s customers and prospects. Others are targeted just at analysts. Many are full of marketing – and all too often, not a lot else.
However, once in a while, I get a real buzz at an event. Such a one was a recent visit to IBM’s Zurich Labs to talk directly with scientists there working on various aspects of storage and other technologies. These sort of events tend to be pretty stripped of the marketing veneer around the actual work, and as much of what is going on tends to be at the forefront of science, then it also forces us to think more deeply about what we are seeing.
Starting with a general overview, IBM then dived deep into flash storage. The main problem with flash storage is that it has a relatively low endurance, as each cell of memory can only be written to a given number of times before it no longer works. IBM has developed a system which can take low-cost consumer-level solid state drives (SSDs) and elevate their performance and endurance to meet enterprise requirements. The team has demonstrated 4.6x more endurance with a specific low-cost SSD model. The impact on flash-storage economics using such an approach will be pretty massive.
However, IBM has to look beyond current technologies, and so is already researching what could take over from flash memory. We were given the first demonstration to non-IBM people of phase change memory (PCM). To date, read/write electronic storage has been carried out on magnetic medium (tape, hard disks) or flash-based media. Read-only storage has also included the likes of CDs and other optical media, where a laser beam is used to change the state of a layer of material from one phase to another (switching between amorphous and crystalline states). For read only, this is fine: once the change is made, it can be left and has a high level of stability. Read/write optical disks need to be able to apply heat to the layer of material back to its base state – and this is where the problems have been when looking at moving the technology through to a more dynamic memory use.
PCM requires that the chosen material can be changed from one state to another very rapidly and back again. It also needs to be stable, and needs to be able to store data over a long period of time. Whereas in optical memory, a laser beam is used, in memory, it has to be carried out through the use of an electronic current. However, there is also a problem called drift. Here, the resistance of the amorphous state rises according to the power law, and this makes the use of PCM in a multi cell configuration (needed to provide enough memory density) a major problem.
IBM demonstrated how far it has got by writing a jpg image to the memory and then reading it back. In a raw form, the picture was heavily corrupted. However, by using intelligent technology developed by IBM, it has created a means of more clearly delineating the levels between the states within the material it is using. Using that system then showed how the image recovered from the memory was near perfect.
Why bother? Multi-Level Cell (MLC) Flash tops out at around 3,000 read/write cycles, but PCM can endure at least 10 million. PCM is also faster than flash and cheaper than RAM: creating a PCM memory system gets closer still to being able to manage large data sets in real time – which many organisations will be willing to pay for.
Next was what being called a “datacentre in a box” (a term I have heard so many times before, that I winced when I heard it). However, on this occasion, it may be closer to being realistic than before. Rather than just trying to increase densities to the highest point, IBM is taking a new architectural approach, using server-on-chip Power architecture systems on a board about the same size as a dual in-line memory module (DIMM), as used in modern PCs. These modules can run a full version of Linux, and can be packed into a 2U unit using a novel form of water cooling. Instead of the cooling being directly on the chip, a copper laminate sheet is laid across the top of the CPU, with the ends of the sheet clamped into large copper bus bars at each end. These bus bars also carry the power required for the systems, so meeting two needs in one design. The aim if for 128 of these modules to be held in a single 2U rack mount chassis, consuming less than 6kW in power. The heat can also be scavenged and used elsewhere when hot water cooling is used. Although “hot water cooling” may sound weird, the core temperature of the CPU only has to be kept below 80°C, so used to cool the CPUs is passed by an external heat exchanger, where its temperature drops to just a low enough temperature to keep the CPU below 80°C before being pumped back round to the CPU. The heat is high grade and can be used for space heating or heating water for use in e.g. washing facilities – so saving on a buildings overall energy bill.
We also saw IBM’s quiet rooms. No – not some Google-esque area where its employees could grab a few moments of sleep, but rooms specifically built to create as ideal a place for nanometer technology experimentation as possible. By “quiet”, IBM is not just looking at how much audible noise is in the room. Sure, there are anechoic chamber elsewhere which have less noise. IBM wanted areas where the noise of electromagnetic forces, of movement, of temperature and humidity could be minimised to such an extent that when they want to hit a molecule with an electron beam from a meter away, they know it will happen.
These rooms were designed not by an architect or a civil engineer. It took one of IBM’s physicists to look at what he really needed from such an environment, and then to come up with a proto-design and talk to others about whether this would be feasible. Being told that it wouldn’t be, he went ahead anyway. These rooms are unique on the planet at this time – and put IBM at the forefront of nanometer research.
These areas we were shown, along with others, raised lots of questions and created discussions that were interesting. As we all agreed as we left the facility – if only all analyst events could be like that.