Post by Jon ElsonThe specs for FS were totally insane, for the technology available
at the time (Motorola 10K ECL or any equivalent). So, should
FS have been canceled as it could NEVER reach the goal, or kept
alive, as it would have been a very powerful machine? Was it
an all-out attempt to make a supercomputer which would sell maybe
less than a dozen units? Or, was it the basis of the next generation
of IBM mainframes?
The 370 series was a practical architecture, although the performance
of some of the lower models seems like it must have been intentionally
crippled to not interfere with the /15x and /16x machine.
re:
http://www.garlic.com/~lynn/2014i.html#40 IBM Programmer Aptitude Test
http://www.garlic.com/~lynn/2014i.html#44 IBM Programmer Aptitude Test
http://www.garlic.com/~lynn/2014i.html#47 IBM Programmer Aptitude Test
http://www.garlic.com/~lynn/2014i.html#48 IBM Programmer Aptitude Test
http://www.garlic.com/~lynn/2014i.html#52 IBM Programmer Aptitude Test
http://www.garlic.com/~lynn/2014i.html#54 IBM Programmer Aptitude Test
http://www.garlic.com/~lynn/2014i.html#55 IBM Programmer Aptitude Test
FS specs had a lot of blue sky ideas ... some of them not even having
any idea about how they might be implemented. since it was suppose to
completely replace 370 ... internal politics during the period was
suspended and/or killing off 370 efforts. some past posts
http://www.garlic.com/~lynn/submain.html#futuresys
some other refs:
Discussion of old FS evaluation
http://www.jfsowa.com/computer/memo125.htm
FS description and discussion
http://people.cs.clemson.edu/~mark/fs.html
wiki entry
http://en.wikipedia.org/wiki/IBM_Future_Systems_project
FS design/architecture was divided into something like 13
sections/areas. My wife worked for head of one of the sections and had
some responsibility for dealing with other sections ... and was
repeatedly surprised/astounded by the lack of any substance backing up
some of their fantasies.
part of FS was sort of object with potentially five levels of
indirection (& storage access; aka an "hardware" ADD instruction which
would handle whether the operands were decimal, floating point, integer,
etc ... or even the same). one of the final nails in FS coffin was study
by the (IBM) Houston science center ... that if a FS machine was made
out of the fastest available hardware ... and an application from
370/195 was moved over to it ... it would have throughput of 370/145
(about factor of 30 times slowdown).
another feature was it was to be "single level store" architecture
... somewhat carried over from tss/360. at the univ. I got to play with
cp67/cms on weekends and sometimes had to share the machine with IBM SE
playing with TSS/360. At one point we did synthetic benchmark for
Fortran edit, compile, link and execute. I got better throughput and
interactive response for 35 simulated users on cp67/cms than he did for
four simulated users on tss/360 (with exact same hardware). I've
periodically claimed that a lot of what i did for cp67/cms paged-mapped
filesystem in the early 60s took into account of "what not to do" from
observing tss/360 (i could easily get three times the native cp67/cms
filesystem throughput). this contributed to my references to
periodically ridiculing the FS efforts (and continued to work on 360 and
then moving to vm370/cms ... during the FS period). posts mentions doing
cp67/cms paged-mapped filesystem
http://www.garlic.com/~lynn/submain.html#mmap
also part of recent discussion over in ibm-main
http://www.garlic.com/~lynn/2014i.html#66 z/OS physical memory usage with multiple copies of same load module at different virtual addresses
http://www.garlic.com/~lynn/2014i.html#67 z/OS physical memory usage with multiple copies of same load module at different virtual addresses
http://www.garlic.com/~lynn/2014i.html#68 z/OS physical memory usage with multiple copies of same load module at different virtual addresses
This goes into major motivation for FS was countermeasure to clone
controllers ... that FS would have such tight integration between
processor and controllers that it would make it extremely difficult for
clones to keep up (but much of the actual specification to accomplish
that was totally lacking)
http://www.ecole.org/en/seances/CM07
other posts mentioning clone controller work
http://www.garlic.com/~lynn/subtopic.html#360pcm
A related subject is the end of ACS/360 (which also gets into tiered
processor performance)
http://people.cs.clemson.edu/~mark/acs_end.html
mentions that it was killed because management was afraid that it would
advance the state of the art too fast and they would loose control of
the market. at the end of above, it goes into some of acs/360 features
finally showing up more than 20yrs later in es/9000.
the person responsible leaves and starts his own clone processor
company. accounts of the lack of 370 products during the FS period is
then credited with giving clone processors a market foothold. This
recent post (in this thread) mentions that it was initially with univ. &
scientific ... before breaking into the true blue commercial market.
http://www.garlic.com/~lynn/2014i.html#52 IBM Programmer Aptitude Test
the folklore is that some of the FS people retreat to Rochester and do
the system/38 ... significantly simplifying a lot of FS features ...
and not having to worry about throughput in the market that they were
selling to. For instance one of the simplifications was that they
treated all connected disks as a common storage pool for single system
filesystem (with any file potentially having scatter allocation across
all available disks). As a result, everything had to be backed up as an
integral whole. A common failure of the time was single disk failure
... but because of the common storage pool paradigm ... the one disk
would be replaced ... and then a complete system restore would be needed
(could easily take 24hrs elapsed time).
http://en.wikipedia.org/wiki/IBM_System/38
and
http://www-03.ibm.com/ibm/history/exhibits/rochester/rochester_4009.html
the followon was as/400 which was replacement for s/34, s/36 and s/38
(and dropped some of the s/38 FS features).
http://en.wikipedia.org/wiki/IBM_System_i
--
virtualization experience starting Jan1968, online at home since Mar1970