Project Monterey was an attempt to unify the fragmented Unix market of the 90s in to a single cross vendor Unix that would run on Intel Itanium (and others). The main collaborators were: IBM who brought its AIX, HP was supposed to bring some bits from HP-UX, Sequent from DYNIX/ptx and SCO from UnixWare. The project shared fate of Itanium – it totally failed. In the end Linux took its spot as a single Unix. The main legacy of Project Monterey was the famous SCO vs IBM lawsuit.
IBM did however produce AIX version for IA64 architecture! According to Wikipedia, 32 copies were sold in 2001. Except of course no one has kept a copy and the famous OS was lost forever.
Until now! This rare release has been recovered, imaged, and uploaded for posterity. It’s going to be difficult to actually run it, though, as there’s no emulator capable of running it – you’re going to need a very specific type of Itanium machine, an Intel Engineering Sample Itanium workstation, which were available from several vendors.
“This rare release has been recovered, imaged, and uploaded for posterity.”
Sadly I have seen corrupted files on archive.org…
AIX is an OS that I really dreaded and complicated the push for standard Unix tooling and application building. As a software vendor, we had to go to great lengths to build and deploy our products on AIX – none of the system administration was standard, and many of the system calls were IBM specific. While there were many other OS’s that required custom work (OSF/1 nee True64, HP/UX, and even SunOS) the differences were less invasive and required less investment to work with. If you had something working on AIX, you had it working on AIX and couldn’t generalize that. Finally, AIX experience didn’t translate to other OS’s – I’m not saying IBM’s administration was awful (others can make that judgement), but it often was very IBM specific, even down to the terminology.
IBM lived in their own computing universe and had a severe allergy to anything not invented by them. So when it came to unix they seem to have gone out of their way to IBMinfy AIX.
We got a few RS/6000 donated to a lab I worked in uni. The sysamdin staff loathed the machines. There was also very few scientific/engineering software so they basically were glorified xterms for the Solaris/Irix that ran most of our SW.
The impression I got from AIX was that it was a very proprietary business-oriented system that was unix because it had to, not because it wanted to. So very IBM.
I’ve got an Itanium system here that’s been sitting for a very long time. I just downloaded these ISO images and they all mounted on loopback on Linux, so that’s a start anyway. The contents of the images definitely look like AIX 5L, which I have used, but there’s nothing like a README or INSTALL file or anything. Anyways, I’m going to see what this Itanium system can do.
dcantrell,
I wanted to play with one back in the day, but it was unattainable. They were so inaccessible to software developers and consiquently itanium would never get the native software needed showcase it’s capabilities. Running software emulated and/or ported from x86 was a bad use of itanium but unfortunately that is what happened.
I ran Oracle on Redhat on Itanium about 2007 or so. It was unstable. We had really long Oracle queries for a federally mandated Medicaid program. They had to run for almost 24 hours. As often as not they would error out. Only a reboot before the query would insure a successful completion. This did not inspire confidence.
AndrewZ,
Interesting, although it raises the question of how it would compare to other machine architectures given similar specs. When I think of SQL queries, it doesn’t seem like a good use of a VLIW architecture, at least not without a radically new approach to databases (*). AI, simulations, and other applications that benefit from long vector operations seem like a better fit.
* Food for thought: how might you use a GPGPU to accelerate database queries? If you rebuilt database primitives from the ground up it might be blazing fast compared to the CPU, but the level of effort to get there would be high. If you merely port conventional algorithms it would be slow and the parallelism would be woefully under utilized.
This particular query was very disk bound. It processed a whole state worth of Medicaid claims for one year. Maybe with today’s RAM and SSD it wouldn’t be quite as disk bound. I have to admit I haven’t looked at SQL optimizations into GPU architecture. Interesting perspective!
IA64 was a superscalar VLIW. Most GPUs are SIMD VLIW.
There are some database kernels that benefit greatly from SIMD. Which is why a lot of enterprise/datacenter infrastructure buys intel and not AMD; AVX512. Although this may change soon with Zen4.
javiercero1,
Theoretically yes, although I couldn’t find any sources confirming this was done for mainstream commercial databases on IA64 or GPUs. Today a lot of software might be compiled for AVX relying on the compiler to auto-vectorize sequential algorithms, but that can leave a lot to be desired because optimization often requires high level changes too.
IA64 doesn’t have AVX4512
javiercero1,
That’s obvious. But databases on IA64 is what this thread was talking about. Did oracle use vector acceleration on itanium? Probably not. I’m not finding much information about oracle queries using vector accelleration on any architecture. But if you find some info, please link it!
It seems only HP-UX and OpenVMS were the only properly supported OS on Itanium.
Linux ran ok, and almost all open source stuff compiled and ran just fine on it. It was just too expensive to really justify using, and the performance was mediocre.