Reader Feedback on AS/400 to i Mystery Solved
February 21, 2011 Hey, TPM
You wrote: “I would love to see a Windows and SQL Server skin for IBM i that would let Windows applications run on the i box, in fact. I am going to think about what that might mean and how one might build such a thing. IBM is using the PowerVM Lx86 emulator to run Linux apps compiled for x86 on Power processors, and perhaps it can use the same QuickTransit tools to emulate Windows apps on IBM i. To do so would merely require IBM to want to try to crush Windows with its own platforms, and IBM doesn’t seem to care about platforms much these days. It cares about smarter planet, several layers of abstraction up from a single server platform.”
I have been thinking about this topic recently and I have would like to propose a radically different approach. Assume that Microsoft would agree to support Windows running on Power, how does this help the IBM i community? Some but not much. But what if the opposite were to happen? What would happen if IBM i could run on Intel servers? I think this option would do more to promote and revive the IBM i community.
That would be the obvious thing to do, wouldn’t it? What I don’t know for sure is how reliable the resulting machine would be and how much work would be involved. The firmware on X64 and Power systems is radically different, but they use many of the same disks, I/O controllers, and other peripherals these days. I think the PowerVM hypervisor is tied very tightly to Power-based iron, and I suspect that IBM has done a lot of work to virtualize the IBM i platform both down in the systems licensed internal code (SLIC) layer of the operating system and in the hypervisor to tune IBM i performance and allow it to make the best use of Power6 and Power7 chips.
Having said that, IBM i is already getting short-sheeted on the CPU thread counts, so all of those threads in the Power7 chips are great for AIX and Linux, but are of dubious use to large AS/400 shops. And it looks like OS/400 and i don’t scale well beyond 32 cores, either. So provided that the RAS features on, say, the high-end Xeon 7500 or Opteron 6100 processors was good enough, you could probably do a port to these chips and get a box that ran pretty well.
What I can tell you for sure is that back when IBM did the transition from CISC AS/400s to PowerPC AS/400s in 1995, the operating system was ported to C++ and now includes a fair amount of Java, which makes IBM i as portable as any other operating system on the planet. (It was a mix of PL/MP, an IBM Rochester variant of PL/1, assembler, Modula-2, and C code in the pre-PowerPC versions of the OS/400 operating system.) There is no reason why IBM i could not be ported to a select few configurations of System x tower and rack servers and BladeCenter blade servers. And as a number of us here at The Four Hundred have said over the years, it is probably a good idea.
But IBM wants to prop up its Power Systems business at all costs. Once you start talking like this, you start wondering why z/OS and AIX can’t all be ported to X64 iron. And why this hasn’t happened already.
If IBM moved all of its operating systems to an X64 platform, there would be a huge application porting issue, as Intel faced with Itanium chips. And the move to X64 platforms would mean that IBM would no longer be able to command higher prices on its iron, which it justifies by saying they have higher RAS and features that fit hand-in-glove with IBM’s own operating systems. A port to System x iron might also open IBM up for antitrust scrutiny if Big Blue didn’t allow the operating systems and their systems stack to run on other non-IBM X64 machines. IBM could be accused of tying its software to its hardware, which is an antitrust no-no that the company has been busted for in the past. (You’ll notice, of course, that Apple gets away with this with its Mac PCs. Apple never had a monopoly. Well, not yet.)
I think IBM would rather do as little work as possible and extract the most money as possible from its server customers, and that means pitching a converged i-AIX backend box and Wintel X64 servers as application and front-end boxes as a fallback position. And yes, I think this is an unimaginative and disappointing strategy considering the architectural benefits that IBM i has.