Reader Feedback On The AS/400 Turns 27, And Still Has Much To Teach IT
June 29, 2015 Hey, TPM:
It’s unreal how often I keep bumping into you. I was intrigued by the title of the article in the attached note, followed these links, and again found an article by you. (And thanks for the reference.) I’ve shared your emotions about IBM and this system, with ever increasing sadness, from the inside.
Still, you laid down a challenge and that reminded me of something I was asked to review by an IBM friend named Phil Vitale. He asked me to review and comment on a paper he had read named The CHERI capability model: Revisiting RISC in an age of risk. As you will see, it was an attempt to define a new processor architecture with a new type of pointer, largely to aid in systems security at its base. Phil thought I’d be intrigued and I was. At that time I provided a review of it by writing up a web page that largely described IBM i’s addressing architecture and did some comparative anatomy. It can be found here.
So, extending this to your challenge.
It strikes me that, by playing with similar concepts, and with some knowledge of the IBM i, and some inventive addressing like that, architects would realize that they are not really all that far from the system that you seem to be suggesting. They already have huge address spaces, some parts of which can be shared exactly between processes. They could add a notion of persistent effective and virtual address ranges. They could include a notion of “commit” on all or portions of objects, ensuring that such objects (or representations of such objects) are copied into physically persistent storage. They could include something like the burst buffers or even locally persistent memory ensuring that the objects are able to be made persistent quickly. I imagine that I could go on for a while, but I suspect you already get much of the picture. The AS/400’s and IBM i’s underpinnings are not really all that magical.
So, as you also seem to suggest, it also astounds me that the many very gifted folks out there can’t seem to pull together an intrinsically secure, implicitly fast, easy to use, massive storage, shared persistent object-based system. (Perhaps they are all playing with Hadoop. More likely they weren’t taught about addressing.)
Do you get feedback from active IBMers when you write these “this is how we grow the AS000” articles? What do they say?
I am all in on how much better the system could be if IBM put some effort into it on the technical side. But there is some very important and likely expensive foundational work that has to be done. I do not think the single level store makes sense any more. (I am very interested in reading a technical debate on the topic of SLS vs private address space for each process.) And RPG and ILE have been far surpassed by C# and .NET. But the founding genius of Glenn Henry to integrate a database and object security into the OS rings true today.
I tried to reach Glenn a while back–he did not respond. I am not sure if he is in all that great of health. Real genius in this system.
IBM says nothing usually. Once in a while I get someone who will call me up and talk, sometimes it is even a general manager of the iSeries or then the Power Systems business, talking off the record and just brainstorming.
IBM needs to place their hardware back into the technical schools just like they did back in the 1970s and 1980s. People may think this is new technology and take an interest to it. I got my start on a System/36 back in 1988 right when the Silverlake made its debut. After a few years, everything was transferred to the 400. From then on I never looked back. Even now we still call it “The Four Hundred!”
My goodness! I’ve been supporting that system that long? I’m just barely starting to call it a System i. I’m going to retire on this baby. By then it will be 33 years of support. What other system will have survived that long whereby you were proficient on it in the early 1990s, gone into a coma, woke up today, and can still support it?
None that I can think of. Well, except IBM and Unisys mainframes.