IBM’s top advocate for a 50-year-old computer memory concept is finally getting the last laugh.
The concept, known in its latest industry iteration as storage-class memory, is that nobody needs disks. Modern solid-state drives are a compromise because they’re inherently all-flash, but they remain configured with all the bottlenecks of standard drives, even when sold as enterprise arrays.
The future’s been arriving any day now for the past few decades. It decrees that computer architectures won’t have drives at all–only processors, I/O, and non-volatile RAM (NVRAM). The Storage Networking Industry Association (SNIA) has been touting this since at least 2008. SNIA members continued promising a bright future, while noting ongoing challenges at the Persistent Memory Summit in January 2017 in San Jose.
Sometime this year, “We’ll see the first notebooks with NVRAM used in conjunction with standard RAM to expand main memory, or, as the industry puts it, as storage class memory (SCM). This is an old idea–remember IBM’s System 38?–given new relevance by NVRAM’s speed, cost, and density,” wrote storage expert Robin Harris on TechRepublic’s sister site ZDNet in December 2016.
SEE: Storage spotlight: SAN, NAS, tape, and all-flash arrays (Tech Pro Research)
What about that System/38? IBM’s minicomputer debuted in 1979 and sold throughout the 1980s. “Mini” was a relative term of the period compared to the physical size and blazing performance of mainframes–after all, the System/38 was still bigger and heavier than your average Harley-Davidson. The System/38 evolved into the AS/400, in turn leading to IBM’s iSeries, System i, and now Power Systems.
Frank Soltis was its designer and the man responsible for its unique memory arrangement. Soltis, a Ph.D. recently retired from Big Blue, still consults for the company and said he’s glad to see the concept moving from an industry island toward mainstream implementation.
“I have a reputation at IBM and elsewhere. I hate disks and have for the last 40 years. I’ve said we’ve got to get rid of those things,” Soltis said. “Why not create a big enough memory address space to cover the entire disk system if you insist on having disks?”
Back then, Soltis called this “one-level store.” He included it when he started building the System/3’s replacement in 1971, which launched as the System/38 by the end of that decade. “That’s what single-level store was all about, that one day we could unplug all these disk drives and just have memory and go from there.”
Unfortunately, the available memory options at that time were not as advanced as today. The System/38 wound up using IBM’s then-standard semiconductor memory, and it’s best known in historical circles as one of the early systems to rely on object storage rather than straightforward files and blocks.
This is a lesson for the modern IBM and other industry titans, such as Hewlett-Packard, Intel, and Microsoft, as they and others all try to figure out storage-class memory.
“Essentially the applications were brand new and people rewrote them for that,” Soltis said. It will happen again in the 2020s, he predicted. This was not insurmountable in the 1980s because of the limited number of applications and because many of the System/38’s customers were new users in general, without much of a legacy systems burden.
Even with the System/38, IBM tacked on a traditional file system shortly after its launch and made it run parallel to the main system due to customer demand. That could happen again if designers aren’t careful.
“Obviously they’ll recompile at the very least, but a number of [applications] may have to be rewritten to move to this new environment,” Soltis warned. “A lot of the people that are looking at the new world will have to find some sort of migration path over time.”