Soltis' views on system architecture and his unique approach to protecting business software assets led to the most successful commercial computer platform in history.
The man known as the father of the System i and today's Power Systems retired from IBM last week after 40 years with the company. Creative, outspoken, and honest about the science he followed in researching ways to perfect the most successful commercial computer platform of all time, Frank Soltis leaves his post as chief scientist for what today is still known in the hallways of Rochester as the AS/400.
"I certainly could not have asked for a better career for the last 40 years," says Soltis. "It's been great.... When I started here in Rochester and was working on the design of what was to become the S/38, I never, never, never anticipated that years later I would be working on essentially the same product."
Soltis says the value of the design that he, Roy Hoffman, and the late Dick Bains came up with is its ability to protect one of businesses' greatest assets: its existing software.
"This is one of the few systems that was really designed for businesses exclusively," Soltis said during an interview with MC Press Online. "It never was intended to be anything else. And the major asset that businesses have--from a computer standpoint--are their applications. As long as you can protect those applications and not force anybody to change, that's the value of it."
One of the unique characteristics of the AS/400, now IBM i running on Power Systems, is its flexibility in allowing changes in the underlying hardware without forcing a rewrite of the applications afterward. Applications that ran on the S/38 years ago still work on Power Systems today. The underlying architecture behind that feature is the Technology Independent Machine Interface (TIMI).
Soltis says the solution was not one he and his team arrived at without trial and error. It was the result of trying to build an interpretive machine and concluding that it would have to be a compiled interface instead.
"I wanted to prototype a high-level-language machine," says Soltis, recounting how the basic architecture of the S/38 came to pass. The idea was in vogue at a number of universities around the country. "In the lab, I was working with a couple of young engineers who put it together for me. I designed a machine, a computer system that directly implemented RPG. Now, at the time, I felt that it should be built in software, not in hardware.... We had an interpretive execution of RPG. That taught me that was not the way to go. The right approach, rather than direct interpretation of this interface, was actually to compile it down into an internal interface--and then execute that.
"That really formed the basis of what became the S/38. I also discovered that picking a high-level language was the wrong approach. What you really needed was a broader interface that could deal with languages, could deal with other utilities--what we call middleware today. That was really the beginning of what we talk about today as MI, or what became known as the Technology Independent Machine Interface."
Soltis says the idea behind objects on the S/38 came from his assignment to work with the Future Systems Task Force, an IBM task force charged with coming up with a replacement for the S/370. The group seemed to wrangle for months, making little progress because of philosophical differences among members, and Soltis eventually was recalled to Rochester due to an imminent restructuring of the company.
"One of the things that was great about that task force," Soltis recalls, "is that there were certain things that we could steal. One of the things that was blatantly stolen from Future Systems was the concept of objects." While mildly dramatic, the comment understates Soltis' prior interest in and work with objects during his studies at Iowa State University while researching virtual memory.
The other unique feature of the S/38 and subsequent platforms is the single-level store, or the idea that all storage on the computer is a single plane of addresses pointing to pages in both primary and secondary storage. Soltis says his underlying concept, however, was that everything in the system could be virtualized, not just storage. Soltis explored and expanded upon the idea in his PhD thesis and then brought the concept back to IBM after earning his degree during an educational leave.
Soltis recalls with nostalgia the work that he, Bains, and Hoffman did on the S/38 and subsequent platforms over the years. He credits Bains with making the platform viable by refining the compiler technology to the point where the system's performance was acceptable, given the limited processing capacity available in the 1970s. That Bains suddenly died of heart failure within just weeks of Soltis' retirement stands as an ironic reminder of how much time has passed since the three men began their career-long quest to perfect what some believe is the ideal business computer.
Asked what he sees coming down the ever-changing road bringing new technology, Soltis reiterates his belief that it includes a new operating system or operating system extensions, or a new programming language or extensions to existing languages. The reason? Rapidly advancing multi-processor technology is making today's programming tools obsolete. It's one thing to try to program applications to run on two or four processors but quite another when there are hundreds or even thousands of processors running at the same time. The current solution of programming threaded applications just isn't going to scale, says Soltis.
"Certainly, the direction we're being encouraged to go today with programming applications that are purely threaded has a limit, and we're going to hit that limit before too long," says Soltis. "I don't have any problem with multi-threaded processors. Since the first one in 1998, we've been using them for some time. But it's not going to scale."
Soltis says the future can be found in the programming approaches taken with super computers and, ironically, computer game consoles that use dissimilar, multiple processors all working together.
"Granted, the Cell chip today is pretty much used for multimedia purposes--special-purpose processors for...video and other kinds of things. But as we go into the future, we're going to see designs similar to that, where the special processors are actually supporting business applications." Functions that are needed to run a database more efficiently or handle TCP/IP in hardware are being handled separately today, and "these [new approaches] have to be reflected in the applications and the way we design them and implement [them]," Soltis says.
Such challenges and the need for new approaches to utilize the extreme power found in tomorrow's processor technology will continue to present opportunities for the next generation of system architects, Soltis says. Anyone who believes everything that can be invented already has been invented isn't looking deep enough into the opportunities that technology will present over the next decade, he says.
Good friends with many in the industry, Soltis intends to continue to keep his hand in the Power Systems world and already has agreed to work with several Business Partners and user groups on projects they have identified for 2009 and beyond. While those familiar with his iconoclastic presentations at COMMON may be lucky enough to hear him at a future event, his larger audience will consist of students at the University of Minnesota, where he plans to intensify his teaching schedule following a period of classroom restrictions imposed by work and travel.
If not in the classroom, however, Soltis says he likely can be found in his garage, working on assembling a collection of classic car parts that heretofore have been cluttering up the place and earning him a reputation for finishing tasks at home that belies the monumental reputation he achieved at IBM nurturing the world's best business platform.