That is IBM i, aka OS/400, a completely different beast than s390. It's one of the most underappreciated and misunderstood examples of systems software in the

That is IBM i, aka OS/400, a completely different beast than s390. It's one of the most underappreciated and misunderstood examples of systems software in the industry. It's survived four distinct CPU changes of the same magnitude as Apple m68k->ppc->ppc64->ix32->ix64, except retaining full compatibility back to the original (and even further back to System/3x). Because everything above the analogue of the syscall layer is machine independent code, you get new benefits with even minor hardware ISA upgrades for existing code without rebuilding. This is all basically "free" for the user without any thought or downside. IBM was also doing ahead of time Java compilation to MI for ~15 years. IIRC this was one of the fastest platforms for JVM apps for many years. The closest equivalent outside OS/400 would be like using LLVM IR for the entire system. But the memory model is also very elegant, the Single Level Store allows for storage tiering without the kind of block layer hiding that most solutions use.. closest equivalent is ZFS with a L2ARC in the POSIX world. The lessons learned from the machine dependent part of IBM i were used to build the pHyp hypervisor in POWER4+. This is a firmware level microkernel and very interesting in its own right. This isn't to say you should go out and port everything to IBM i, just that it should be studied by any serious OS developer and academic. While there is a good book on the design by Frank Soltis, the chief architect, it does not go into anywhere near the depth that UNIX design books did even when covering the proprietary versions. I wish I could learn a lot more about it.