7 Mar
2017
7 Mar
'17
1:50 a.m.
On Mon, Mar 6, 2017 at 5:59 PM, Clem Cole <clemc(a)ccc.com> wrote:
But at that time, LCC was putting things in AIX,
Ultrix, Tru64, HP/UX,
DG/UX, Prime-ux, Intel, AT&T and a host of others. It was kinda
neat setting everyone's dirty laundry though.. you learned a lot.
I left out Solaris and SunOS too.
And BTW: Being lucky enough to have hacked on the kernel of almost all of
the majors at one time or another, I've been asked an interesting
question. Which was the best to work on.. they were all different is the
best answer, I give.
But the fact is that DG was a source licensee but they did a full kernel
rewrite starting I want to say in the late 1980s, early 1990's to build a
scalable SMP. It was probably the easiest of all the kernels I ever got to
hack on. Very clean, well documented and the locks were easy to
understand. We did a study for DG to TNC into it, be they never pulled the
trigger. We quoted it faster than any other port, because our experience
had been that everything we had done on DG/UX had gone so smoothly. But
we'll never know. They died shortly after the study was finished, which
was a shame. I've sometimes wonder what happened to that IP.
It would be interesting to compare it to OSF/1, which was probably the
other very cool kernel I hacked on extensively for both Intel and later DEC
of course. DG/UX was not quite as modern as Mach from a standpoint of
things likes "ports" or being a uKernel - but as a pure well documented and
easy to understand SMP UNIX kernel it was hard to beat.
I did do a little work with Chorus and still have the doc set, but never
worked with enough to have an opinion of how good it was. It showed
promise and I know the UI/AT&T guys had hoped to go there at some point.
Larry's described Solarius pretty well and my experience match his, but I
always thought that the locks were madness IMO - so easy to get wrong.
SunOS was a lot simpler and as Larry has said was pretty elegant for what
was there.
HP/UX was pretty darn bullet proof. The HP folks worked on fault tolerance
got rid of panics more than other other UNIX we say, which was pretty
amazing, but it was not the easiest kernel to mess with. We did manage
to splice the vproc layer and TNC in it and we had a lot of fun with
process migration. Its too bad that never shipped. Again, I've wonder
about that IP too.
Clem