The AS/400 At 28: A HENRY, Not A DINK
June 20, 2016 Timothy Prickett Morgan
A lot of people get credit for the work that went into creating IBM midrange systems over the decades, and rightfully so since the creation of such machinery, including the hardware and the software, takes many hundreds or thousands of individuals with each successive generation. It is appropriate to think about them as the AS/400, the forebear of the current Power Systems IBM i system, turns 28.
That is a long time in a human life, and an even longer time in computer history, where a generation has spanned from 12 to 24 months instead of decades. The funny bit is that the span of a computing generation is stretching out to maybe more like 30 months these days as Moore’s Law slows, and the younger generation is also stretching the time between when they are born and they bring the next generation into the world. It is not at all unusual for kids these days to be 30 to 35 years old before they settle down and grow up. (Funny juxtaposition of phrases there.) In fact, I have heard a new abbreviation for this generation: HENRY, short for High Earners, No Responsibilities Yet. The yet is the key there, and it is what separates these people from DINKs, which means Double Income. No Kids and implies that this will be the way it is for these particular couples.
The generations stretch over time, and we have to go back from time to time to learn–and often relearn–what others did. Each year, as the AS/400’s June 21, 1988, birthday rolls around, I get to thinking about the evolution of computing and how strange it is that this machine exists at all, what a success it has been in terms of innovation and market acceptance over the years, and also how its predecessors and follow-ons have come to embody a certain way of thinking that is both elegant and clever (as if you could have one without the other), and that equally importantly still brings value to the 125,000 or so businesses that employ it as the main system.
As I write this, we are also approaching Father’s Day, so it is also hard not to think about the systems that led to the AS/400, significantly the System/3 from 1969 and the System/38 from 1978. While Frank Soltis, the long-time chief architect of the AS/400 system and one of the key designers of the System/38, often gets a lot of credit for its development, he was one of many who fought for the machine and helped keep it alive. One of my favorite people in the IT industry, Glenn Henry, who created clone X86 processors after leaving IBM in the 1990s with a company called Centaur Technology because he thought Intel’s processor prices were too high, was instrumental in the development of the System/38 as well. I wanted to give him a shout out as well as to the good Doctor Frank.
If you want to read a fascinating interview, check out Henry’s interview from 2001 that is part of an oral history of the computer business put together by the National Science Foundation and housed at the University of Michigan. This is how Henry responded when asked about the thing that made him most proud in his illustrious career:
“The thing I’ve done that I am most proud of is the System 38 and I’ve done lots of things since then, but this is a very innovative system done by a bunch of young kids in the cornfields of Rochester, which was a backwater of IBM, and the system’s key concepts are modern today, the system has survived today. I am really proud of the team. I’m proud of what I did there. Yes, so it took seven years to develop it, I mean, but it represented at the time, in fact at one point I made a chart, you know typical ACM paper, the 10 great advances of computer science–I made a list–we had seven of those on our system. You know, tags, storage, single level storage, integrated database, you know those types of things. We had built these into a commercial system and made it work and delivered it to customers, right, with little help from anybody. Research didn’t help us at all. In fact, I have a very bad relationship with IBM Research, and it started here. They didn’t support us. You would think that they would have but they didn’t. The forces of the mainframe people were constantly against us, constantly chirping, trying to kill us because we were an annoyance to them. You know IBM was very competitive. They owned MVS and 360’s and we were building this thing. They just saw it as some SkunkWorks project that would perhaps interfere with their grand scheme of things.”
This cracks me up. As I said last year around this time, what does an in-memory database mean in a world that doesn’t have single-level storage? It means precisely that: Your machine’s architecture was so short-sighted that it didn’t have single level storage at all! The idea of asymmetric processing is also coming back, we just called it accelerated computing now. But it is still the same idea: Offload certain functions from the CPU to intelligent controllers lashed to the main memory of the CPU, and make it all look like a single collective as far as applications are concerned.
We are glad that Soltis, Glenn, and so many others fought the battles inside of IBM that allowed the System/38 to come into being and kept pushing so that a decade later the AS/400 made the System/38’s core technology inexpensive enough that IBM could offer integrated relational database processing at the price of a System/36 that only had flat file databases. We just wish that the Power Systems-IBM i combination today was extended from systems of record to systems of engagement, the ones where advanced data analytics (which is driving an increasing part of the workload in the data center) is done.
IBM has to figure out a way to move Spark, Hadoop, and other workloads on to the IBM i platform itself, much as it has done with its System z mainframes, so they can run applications created for Spark and Hadoop even though these are very different from the architecture of the single-node, multicore Power Systems iron that most customers have. I am not sure how to do this, but the machine can be partitioned and the complexity of the virtual cluster masked from end users. IBM needs to make sophisticated analytics that merges structured and unstructured data as invisible in 2016 as it made the relational database with the System/38 and perfected with the AS/400.
I know that IBM has no stomach for such things any more, but what I really want the company to do now that it has fully committed to the Power processor and platforms that are based on it is to create a next-generation AS/400 that takes the ideas of the system and applies them to modern, scale-out architectures while preserving (as much as possible) the single-level storage and asymmetric processing genius of the AS/400. I suppose to a certain way of thinking of it, a Power9 chip with NVLink and CAPI ports talking to GPUs and other kinds of accelerators looks a bit like an AS/400 CISC processor with a few dozen or hundred Motorola 68K I/O processors hanging off of it to offload various kinds of processing from the CPU. The virtual and unified memory features that are going to be used in supercomputers based on Power9 chips and Nvidia Tesla GPU coprocessors are a kind of single level storage, we suppose. We would just like to see a world where IBM i is a part of this, or that IBM would make Linux more like IBM i and therefore be able to run RPG, COBOL, Java, and PHP applications on the same iron that does simulation or analytics jobs.
In short, we want another generational change, one that embodies all the best ideas of today and that takes the IBM i platform into new realms and into the next decade or two. We have no idea what kind of software investment this might take, of course, or if customers would go for it. But we need a new Future Systems effort from IBM and its OpenPower partners that includes IBM i rather than just shunning it. This is a lot to ask in 2016, we realize. But it sure would be fun to show that Google doesn’t invent everything these days, and that you can, in fact, create a single platform that is good as a system of record and a system of engagement. That is, in fact, an Application System. Period. And that shows it was stupid to divide them up in the first place.