Wampum

I spent the years 1984/1985 working at an ICL facility in Bracknell, and at a Siemens facility in Munich. I was writing the European Unix Specification, mostly by myself, spending nights reading my way through about 20 variations on the v7, Xenix, System III and 4bsd implementations that were offered by my client -- Bull (France), ICL (UK), Siemens (W. Germany), Olivetti (Italy) and Nixdorff (W. Germany) -- to find "vanilla", the comon computational surface they could provid, at acceptable engineering cost. Every system call, every library routine, every file format, every logical device.

I made one mistake. Nixdorff was selling Sequent's multi-processor, and wanted to shave bits off of the PID (process id) to identify which processor a process was bound to. For two or four processor SMP systems this was just one or two -- or three if the challenge of an eight-way system was attempted -- bits subtracted from a 16 bit object, and at run-time the system would have to roll-over the process ids after 16,384 or 8,192 or (worst case) 4,096 processes had been created and ended ...

Of course, if anyone ever built an SMP with say, lots of processors, process id roll-over would take place before the system completed the boot sequence. Here's a note I sent to the Linux kernel list -- note.

The point of this trip down memory lane isn't that I wrote an external or programmer's specification of an operating system, I've written more than one of those. It is this: compiled languages like C abstract away the specifics of the underlying instruction set (processor architecture), and make programmers fungible. Where there was a market of assembly language programmers for Remington-Rand or Sperry-Univac or IBM or ... there was a market for FORTRAN or ALGOL or ... C programmers. Ultimately, as the Unix and Linux projects manifest, operating systems (as applications of processor architects) become economially processor-independent.

Deep breath: All systems programmers are therefor C programmers, or in really wierd niche markets (things that go "bang" or "buzz") or unemployed.

I spent the years 1992/1993 working at the Locus facility in Los Angeles. I was writing Spec 1170, which is now known as the Single Unix Specification with Greg Thiel and another guy, approaching the problem from the question "which APIs actually are used (and by whom)?" This time my client was IBM and HP. The number of thingiees specified was ... 1170.

One afternoon over a quiet reflective beer in Palo Alto it came to me. The European thing was a rear-guard action to defend ECMA, the European Computer Manufacturing Association, from IBM. I knew that. The Unix Standards Wars had come and gone and come again, each time in a new costume -- Network File System (Sun et al) vs Andrew File System (IBM et al), pitting local-area network and insecure (connectionless) semantics against wide-area network and secure (connection) semantics, won by NFS, and so on. What was now going on was a rear-guard action to defend everybody except Microsoft, and IBM was now one of the market hemopheliacs. We, with all our diversity of vendors and extensions from the comon model, were being forced off of the general desktop and onto the raised floors. We'd lost the sea of glass and were restricted to islands populated by file servers.

First Deep breath: All programmers are therefor Windows programmers, or in really wierd niche markets (things that go "backend" or "bang/buzz") or unemployed. Mostly unemployed, and trivially outsourced.

Second Deep breath: Operating systems monoculture would event produce instruction set monoculture. We'd lost the Sea of Glass to Microsoft, and in time there would be no economic justification for anyone other than Intel to be making commodity silicon.

So the news that the plan of record at Apple is now to use Intel processors comes to me 11 years after I reasoned that it would. For some things I really do have a working crystal ball.

MB allowed me to specify a Mac for the kids. A nice gray iBook, and OS X.4. And to save space I'm converting the Windows 98 partition on my dual-boot (intel) laptop to ... just another Unix partition.