The IBM Edge2014 conference held May 19-23 in Las Vegas showcased a variety of technologies being developed by IBM that promise to be the future of enterprise data storage. The technologies were presented by Clod Barrera, chief technical strategist for IBM System Storage, and Dr. Axel Koester, executive IT specialist for IBM System Storage. Tony Pearson, Master Inventor and Senior IT Specialist for IBM System Storage, has provided notes and information about the presentation for non-attendees.
Some of the research projects show promise of being shippable in products that may wind up in a data center near you, while others don't seem likely to appear in any product available to corporate buyers or individual customers.
A 120 PB file system solution
Using a grid of IBM Power 775 servers and a staggering 200,000 disks, a 120 PB file system run on General Parallel File System (GPFS) is being developed for a supercomputer deployment for an unnamed organization that will perform simulations of real-world phenomena. GPFS Native RAID offers an 8+3 Reed-Solomon data protection scheme, which is an impressively durable and fault-resistant arrangement.
This news is somewhat old to be trotting out at a technology presentation, but this is easily one of the largest storage arrays to be built to date. This was described on launch as being "a system that should not lose any data for a million years," a claim that — as a product of mechanics — is marketing bluster.
IBM is also working on a design that allows users to leverage multiple cloud vendors such as Amazon, Google, and Microsoft. Users would be able to access storage from multiple providers without being constrained by the bottleneck of a single master scheduler.
A phase-change memory card
IBM's prototype phase-change memory (PCM) card uses a combination of PCM and standard NAND. PCM relies on a chemical state change (amorphous to crystalline) in a chalcogenide alloy, with the resistance of the two signaling either a 1 or a 0. Writing to PCM cells is performed by controlled heating, and PCM cells can be written to at least one million times.
According to IBM, the prototype card achieved 65k read IOPS at 35 microseconds, and 15k write IOPS at 61 microseconds. The card also has consistent low latency, as 99.9% of the requests completed within 240 microseconds. This is 12x lower than multi-level cell (MLC) and 275x lower than triple-level cell (TLC) flash SSDs. The highest observed latency was 2 milliseconds, 7x lower than MLC and 61x lower than TLC flash SSDs.
The potential for this prototype to ship is questionable, due to the cost of components — PCM is less expensive than DRAM, but more expensive than NAND. An appropriate use case, or as part of a tiered storage arrangement, could be a promising reason for such technology to hit mass production.
Liquid-state storage is a very unique concept: using an ionic liquid-based transistor, data can be stored. IBM has alternatingly referred to this as "liquid metal storage," which is something of a misnomer (or, at least, misleading); the liquid is vanadium dioxide, not a substance like mercury or gallium, which are true liquid metals. With vanadium dioxide, researchers can give a positive or negative charge, which it maintains (being itself an insulator). This state can be read back, as any transistor, as a form of data storage.
Liquid-state storage is a project of IBM Fellow Dr. Stuart Parkin, one of the inventors of racetrack memory. He notes that, "Using these devices and concepts in novel three-dimensional architectures could prevent the information technology industry from hitting a technology brick wall." Racetrack memory, another IBM project, has yet to materialize into a commercially available product, but some difficulties in development have prevented that from seeing mass commercialization. Perhaps it is in an abundance of caution that it is not yet available, for fear that racetrack memory might become a flop right out of the gate. As for liquid state storage, although there is no promise of it shipping anytime soon, it may be relegated to vaporware status.
Tell us what you think
What do you think the chances are that these projects will ever be in use in your data center? Let us know in the comments section.
James Sanders is a Java programmer specializing in software as a service and thin client design, and virtualizing legacy programs for modern hardware.