[TUHS] Ancient unixes

Jose R Valverde txomsy at yahoo.es
Mon Oct 10 19:37:24 AEST 2005


Waddayamean?

I mean: what does it mean to you 'the spirit of
ancient 
Unix'?

If by that you mean the fact that they were simple, 
slim and efficient, doing one simple thing and doing 
it right, you may then consider the effort by 
ast in the 80's with MINIX. OK, it used it's own 
microkernel, but the basic idea is the same... and has

been followed on by Mach, BSD-lites, Flex, MacOS X,
Tru64, Linux on L4, etc...

As a matter of fact I always felt UNIX after v7 got it

wrong: e.g. network data is no longer another stream 
(I'd have loved it to be a file system with
directories 
representing network addresses and ports being files
or 
pipes). Thus, later unices increased complexity by 
abandoning the simlicity of the original design. If 
that is the case, Plan 9 is a good update. And so is 
Inferno.

Actually, I always felt that many additions to UNIX 
might have been better implemented outside the kernel 
if only the kernel had been expanded to allow
user-mode 
expansions. But that's already here with kernel
modules 
in Linux, BSDs, Solaris, etc... which are becoming
more 
and more microkernelized each day. As microkernels 
become bigger :-)

OTOH, if you mean adding 'modern' services, perhaps
QNX 
is doing it with its support for Real-time. Or adding 
dynamic libraries, networking, modern virtual memory 
(beyond swapping), etc... which at the plainest level 
is what more or less likeably all modern UNIX have 
done.

Extending into the future? Distributed computing, 
clusters, etc? Like some commercial UNIX, Amoeba, 
Inferno and the like?

If you only mean resurrecting these ancient UNIX on 
modern hardware, there have been initiatives to
rewrite 
v7 alike systems for other architectures (say OMU,
UZI, 
MINIX, Coherent, Xinu, etc.). But for that you already

have emulators that provide you the original flavor at

even higher speeds in a virtualized environment.

So? waddayamean?

I think the answer to your question is YES! Lots of 
people have tried to improve ancient UNIX more or less

successfully, and many people is still trying, using 
microkernels, no-kernels, adding RT, VM, dynamic 
libraries, kernel modules, etc... Each with their own 
approach.

This said, if I were to pick an initiative that gets 
closest to the wishes of the original designers, that 
should undoubtedly be Plan 9 and its successor, 
Inferno, as they are what the 'Original Designers' 
themselves have done when they tried to repeat it
doing 
it 'right' (or at least better) no matter what my 
personal opinions regarding the issue may be.

Regarding my opinion, yes, I would go for the good old

leather-bound days of IBM mainframes with MVS. (zOS?) 
which oddly enough are finally reaching the rest of us

with Xen and emulators like QEMU. If I were to wish, 
I'd like a no-kernel approach (everything independent,

cooperating, hot-substitutable, fully migratable 
processes) over a virtualizing system that allows me
to 
run several OSs and update/change any OS component on 
the fly without service interruption, and to migrate 
everything between machines on demand ('cos of
overload 
or hw failures or whatever, or just 'cos I wish to). 
Now, _that_ would IMHO be close to ultimate OS design:

something that can always be updated on the fly and
may 
survive any change, something that can adapt and
evolve 
without interruption or even the user noticing. But 
that is a complex enough concept to expect most system

programmers to grasp, let alone sysadmins, programmers

or users not to pervert. Not to talk of salesmen and 
marketroids!

                           j
--
Jose R. Valverde
EMBnet/CNB



		
______________________________________________ 
Renovamos el Correo Yahoo! 
Nuevos servicios, más seguridad 
http://correo.yahoo.es



More information about the TUHS mailing list