The Soul of Barb's New Machine was creat 1147

Don't take this too personal. It is quite common here in a.f.c.

There were lots of efforts in "de-suckering" unix at the time. The first pbuttes ended in SVR4 and BSD4.2; and it all went into the corporate spiral after that. At one point around 1990 it looked very much like the minicomputer lock-in was about to return.

difficult to navigate between Sun, SGI and others. Software quality often gave a huge stink as well. Irix or SunOS 5.0 anyone?

September 1994 was the first time off-the-shelf unix software really worked for a plain hardware platform. (it may have been 2-3 months earlier, but it was definatly not there in May 1994). This was when debugging and packaging was good enough to do an off-the-shelf install on commodity hardware.

And then there suddenly were lots of alternatives. Linux was just one of many, but got the media attention. I think a major part of this was that unix had gotten the name tainted, and noone trusted Sun with Solaris-x86.

In between there has been 10 years. Linux had sort-of acceptable SMP in 2.4, and it is excellent in 2.6. I am just amazed at the ability to master load on 8+ way machines with tons of disks.

FreeBSD was somewhat-less-stellar in 4.x, but it was still way beyond what Tops10, Multics or Primos ever made it to. There has been a huge improvement in 5.x; although I would rate Linux 2.6 a little better. BSD's seem to have a smaller footprint through; this is visible in FreeBSD, obvious on NetBSD and extreme in OpenBSD. I doubt there are any large lock-out sections of code left in FreeBSD 5.x or Linux 2.6. I know they are there in OpenBSD, and there are probably some left in NetBSD.

It has trickeled over to NetBSD also, but this distribution has given priority to support of oodles of hardware. Even OpenBSD 3.6 has sort-of SMP; and they have kept lean-and-mean to the extreme, with a need-to-have policy all over the place. But even this what I call "sort-of-SMP" would beat Primos hands down, and would give Tops10 a real run for the money.

To the original question; to just bring "job-user state" along, it would require a huge redesign of most systems. It would not be feasible at all in Linux or Windows; there are just too many libraries with internal state.

A microkernel done like QNX could possibly do this, but would require a huge redesign. As soon as you involve network connections the paradigm falls flat though, because you propagate state throughut the world.

It may be possible to bring authenticication or session identification along though, and let it be a pointer to state stored elsewhere.

-- mrr