[email protected] (Paul Gilmartin) writes:
> Further, with UNIX shared memory and the older CSA and LPA
> a single page, possibly executable, can be mapped into multiple
> address spaces.  It better be reentrant.
>
> The address ranges wouldn't need to be identical if the code
> were location independent.  This may have been a goal of the
> S/360 designers, never exploited in OS/360 software.

os/360 relocation adcons became fixed when executable was loaded into
storage before execution.

tss/360 did implement location independent executable images (where same
image could be mapped simultaneously into different virtual address
spaces at different locations.

when i did page-mapped filesystem for (cp67/)cms in the early 70s, I
also implemented location independent support ... but CMS primarily used
OS/360 assembler/compile/loader conventions (with relocatable adcons
that got fixed at load time, rather than the tss/360 model) ... I
frequently had to significantly massage os/360 "generated" code to make
it location independent. I then ported the support to vm370/cms ... and
a very small subset was included in vm370 release 3, w/o the paged map
filesystem and the location independent support.

past posts mentioning horrible contortions I sometimes had to resort
to in order to have location independent code.
http://www.garlic.com/~lynn/submain.html#adcon

I would say that when I did the CMS page-mapped filesystem ... I avoided
doing all the things that I saw tss/360 had done wrong (from performance
stand point, it did do the location independent correctly). Possibly one
of the reasons that the full page-mapped stuff wasn't included in vm370
release 3 was because the (failed) FS effort had pretty much adopted the
tss/360 model ... and "page-mapped" got such a bad reputation with
the FS failure ... some past posts
http://www.garlic.com/~lynn/submain.html#futuresys

even tho I could show my cms page-mapped filesystem had 3-4 times the
throughput of the native cms filesystem (both original and EDF). some
past filesystem
http://www.garlic.com/~lynn/submain.html#mmap

other trivia, something similar also shows up at the time of decision to
migrate all 370 to virtual memory ... recent post mentioning major
motivation was the horrible MVT storage management.
http://www.garlic.com/~lynn/2016h.html#98 A Christmassy PL/I tale

Simpson (from HASP), rather than going to gburg as part of HASP/JES
group ... went to Hawthorne and did "RASP" ... a MFT2 based
implementation supporting an (os/360 oriented) paged-map filesystem
... which showed some of the advantages for (os/360 based) virtual
memory systems.

When it wasn't picked up, he left IBM and joined a clone processor maker
... where he re-implemented RASP from scratch.

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

----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to [email protected] with the message: INFO IBM-MAIN

Reply via email to