Doctor Frank Talks Power With Vision
January 23, 2017 Alex Woodie
Vision Solutions landed a pretty popular guest for the inaugural episode of its new PowerTalk podcast last month. Former IBM chief scientist Doctor Frank Soltis, the Father of the AS/400, shared stories about the sudden popularity of memory-oriented architectures, the limitations of the System/38, and the origins of the MIMIX product.
Becky Hjellming, Vision’s director of product strategy, was excited to have such a distinguished IBMer on her very first PowerTalk podcast. She started out by asking Soltis about his recent COMMON Europe tour, during which he made several predictions about how the computing world is about to change in a major way.
“I’ve been very fortunate over the years in that I’ve spent an awful lot of time in the future, as it were, looking at future technologies and trying to make predictions as to where the world was going, especially when you’re talking about the business computing world,” says Soltis, who retired from IBM in 2008.
The decline of Moore’s Law, in particular, is forcing computer architects to reconsider how they arrange disks, memory, and processors inside of systems. Most operating systems today are disk-based, and therefore have the overhead associated with moving data from disk to main memory to cache memory and finally to the processor itself where it can be used.
That is changing, says Soltis, who was one of the key architects of the single-level storage system that debuted with the System/38 in the early 1980s, which would be carried forward with the debut of the AS/400 in 1988 and of course today’s IBM i running on Power Systems servers.
“The approach we had when we first designed the System/38, we said what if there’s no disk? What if there’s only memory? And by the way, what if that memory is so fast that the processor can operate directly out of it?” he says. “We would eliminate that hierarchy and we would just have a single level of memory. And we refer to that as single-level storage.”
Thanks to the decline or Moore’s Law, other computer makers are “rediscovering” the wisdom of that approach, Soltis says. “Just about every major computer manufacturer in the world is starting to talk about memory-oriented computing,” Soltis tells Hjellming. “One of my favorites is HPE. They’ve been talking about it here for the last several years. They have something they refer to as Universal Memory. Their name, Universal Memory, is what we call single-level storage.”
Soltis is puzzled why it took so long for the rest of the industry to go down the in-memory path. “When we were first doing this design for single-level storage, we all assumed that the entire world was going to be moving there sometime in the 1970s or the 1980s. We’d never believe it would take something like 40 years before people were talking about memory-oriented architectures. But indeed, that’s pretty much what’s happened.”
There are a number of benefits to having a single-level storage architecture, and one of them involves ensuring the resiliency of data. As Soltis explains, the disk in a single-level storage architecture such as IBM i’s is, at a simplistic level, a backup of the data that sits in memory. But you don’t need to wait until the data is backed up–i.e., before it’s written to disk–to do something useful with the data. This brings advantages when it comes to replicating data, which is obviously something that’s of great interest to Hjellming and her Vision colleagues.
“When we’re taking about being able to journal something to another system, we need to do that so we actually don’t lose part of the transaction,” Soltis says. “In other words, we don’t wait until it’s actually pushed out to the disk, to be able to replicate it to a different system.”
While high availability is widely accepted as a necessity these days–particularly for large global companies that operate 24 hours a day, seven days a week–that concept was practically foreign to System/38 customers, Soltis explains.
“At that particular time, the customers [were] primarily small and midsize businesses. The whole concept of 24-hour operations really didn’t exist,” Soltis says. “In other words, businesses would open up in morning. They’d turn their systems on in the morning. Then at night they’d do a tape backup and turn the system off.”
Bigger customers were encouraged to buy IBM mainframes, which were manufactured by a different division at IBM. Big Blue famously encouraged competition between its different divisions, but you could say that it wasn’t really a fair fight.
“One of the things that many people perhaps don’t understand about the early days was, when we first did the Systems/38, the [IBM] corporation limited us to how big a system we could build,” he says. “The problem at the time they were looking at is they did not want any of the System/38s to overlap any of the mainframe space. So as a result, we were to build no systems larger than smallest mainframes.”
That limitation was removed with the launch of the AS/400, when Soltis and his Rochester colleagues were free to build systems as big as they wanted–or as big as their customers wanted, anyway. As System/38 shops migrated to the AS/400s, and as bigger companies started running AS/400s, the high availability requirements began to change, too.
Soltis liked the “continuous availability” capabilities that the mainframe division was building, and he wanted to build something similar into the AS/400. “So we put together a plan,” he tells Hjellming. “We went to IBM headquarters and said we need some development dollars to be able to do this. Well, it turned out we were turned down. Back in those days, in the late 1980s and early 1990s, IBM was having some major financial problems.”
So instead of building the availability technology in Rochester, Armonk told Soltis and his team to find a business partner who would be willing to build it instead.
“I started to look around and see if I could find anybody who could help us with that,” he says. “In fact I found a business in Antwerp, Belgium and it was a shipping company–one of the companies that deals with container ships. Because of the fact they were working with container ships, they had to have their systems–they were using an AS/400–available 24 hours a day, seven day a week, because they had container ships all over the world.”
The company had actually designed its own logical high availability product that used journaling to replicate data between two systems. “I was very impressed by it,” he says. “I tried to get IBM to market it. IBM wasn’t terribly interested in marketing it, so I went to look for business partners. One of those business partners was Lakeview Technology, and Lakeview eventually acquired this product from the Belgium shipping company. That’s sort of the basics of today’s MIMIX.”
You can listen to Hjellming’s entire PowerTalk episode featuring the legendary Doctor Frank Soltis at www.visionsolutions.com/resources/power-talk-podcast-series. It’s certainly worth listening to the whole 30-minute chat. Hjellming has set a high bar for herself with the inaugural podcast. But if she can deliver this sort of IBM i insight every month, then PowerTalk will quickly become a must-hear podcast for the IBM i community.