Groups | Search | Server Info | Keyboard shortcuts | Login | Register

Groups > comp.unix.misc > #76

Re: which one came first

Cross-posted to 2 groups.

Show all headers | View raw

Barry Margolin <barmar@alum.mit.edu> writes:
> If you're going to count non-Unix systems, Multics had memory-mapped 
> files back in the 60's.  In fact, that was the ONLY way to access files 
> on Multics -- the equivalents of open(), read() and write() were in a 
> library that used memory mapping.
>
> Multics was the system that inspired Unix, but they didn't copy this 
> aspect, I think because the early Unix hardware didn't have sufficient 
> virtual memory features.

tss/360 ("official" system for 360/67) was all memory mapped files
... but wasn't very smart about it ... just mapped and then lots of 4k
page faults. along with other inefficiencies ... tss/360 ran a four user
fortran compile & execute slower than cp67/cms ran equivalent 30 user
fortran compile & execute (on identical hardware). tss/360 never did get
very far and was decommitted ... although a few customers did managed to
keep in alive and justify tss/370. Later in the early 80s, there was a
special project for AT&T to take a stripped down tss/370 kernel (SSUP)
that would have unix layer on top.

in the first half of the 70s, company did Future System effort ... which
was going to replace 360/370 ... and was suppose to be all memory mapped.

i was at the science center on 4th flr (multics was on the 5th flr) and
did a memory mapped file system for cms. I had learned quite a bit what
"not to do" watching tss/360 ... I also figured that i could also do as
well as people on the 5th flr (as well as critical of future system not
really knowing what they were doing ... drawing analogy with long
playing cult film down at central sq). misc. past posts mentioning
545 tech sq
http://www.garlic.com/~lynn/subtopic.html#545tech

misc. past posts doing cms memory mapped filesystem ... was deployed
internally, but didn't make it into product. 
http://www.garlic.com/~lynn/submain.html#mmap

I could get 3 times the thruput (of standard cms filesystem) for
moderate i/o application. part of the issue was cms running in virtual
machine used channel programs for i/o (including file i/o). 360/370
channel programs were "real" ... so channel programs created in virtual
machine had to be "copied" into real channel programs with real
addresses. Going to memory mapped filesystem allowed me to eliminate a
whole layer of emulation and enabled a doing a lot of optimization (that
wasn't possible with a channel program emulation paradigm).

however in the wake of both tss/360 and future system memory map issues
... there was lots of resistance to memory map paradigm (even if there
was lots of internal sites running it and showing improved performance).

-- 
virtualization experience starting Jan1968, online at home since Mar1970

Back to comp.unix.misc | Previous | Next — Previous in thread | Next in thread | Find similar

Thread

which one came first jgk@panix.com (Joe keane) - 2012-04-23 19:05 +0000
  Re: which one came first David Dyer-Bennet <dd-b@dd-b.net> - 2012-04-23 14:32 -0500
    Re: which one came first Barry Margolin <barmar@alum.mit.edu> - 2012-04-23 17:56 -0400
      Re: which one came first Anne & Lynn Wheeler <lynn@garlic.com> - 2012-04-23 18:50 -0400
        Re: which one came first Anne & Lynn Wheeler <lynn@garlic.com> - 2012-04-23 21:49 -0400
  Re: which one came first scott@slp53.sl.home (Scott Lurndal) - 2012-04-23 21:44 +0000
    Re: which one came first jeffj@panix.com (Jeff Jonas) - 2012-05-15 02:41 -0400
      Re: which one came first sarr.blumson@alum.dartmouth.org - 2012-05-15 13:45 +0000
        Re: which one came first Barry Margolin <barmar@alum.mit.edu> - 2012-05-15 10:14 -0400
        Re: which one came first Ahem A Rivet's Shot <steveo@eircom.net> - 2012-05-15 15:29 +0100
          Re: which one came first "Charles Richmond" <numerist@aquaporin4.com> - 2012-05-17 09:28 -0500
            Re: which one came first Casper H.S. Dik <Casper.Dik@OrSPaMcle.COM> - 2012-05-17 15:24 +0000
              Re: vfork and stuff which one came first John Levine <johnl@iecc.com> - 2012-05-17 22:35 +0000
            Re: which one came first Ahem A Rivet's Shot <steveo@eircom.net> - 2012-05-17 16:39 +0100
  Re: which one came first Peter Flass <Peter_Flass@Yahoo.com> - 2012-04-23 21:07 -0400
    Re: which one came first Peter Flass <Peter_Flass@Yahoo.com> - 2012-04-23 21:12 -0400

csiph-web