Re: What would be a truly relational operating system ?

From: paul c <toledobythesea_at_oohay.ac>
Date: Sat, 14 Nov 2009 00:15:02 GMT
Message-ID: <aEmLm.51903$Db2.3545_at_edtnps83>



Anne & Lynn Wheeler wrote:
> paul c <toledobythesea_at_oohay.ac> writes:
>> I remember reading an Datamation interview of Gene Amdahl, long ago,
>> must have been in the 1970's because that magazine was one of the few
>> trade mags then.  I still remember it because he was waxing on about
>> his 1950's designs and what came to be called complex instruction
>> sets.  He was regretting that there was little industry enthusiasm for
>> even more complex instructions, I got the impression that he had felt
>> he was barely scratching the surface with ones such as 'edit and mark'
>> or 'translate and test'.  If I had to pick just one target for
>> applying complex instructions, that would be something like the D&D
>> A-algebra.
>> (This wouldn't prevent parallelism under the covers.)  Unlike most of
>> us Gene Amdahl is able to visualize approaches that are contrary to
>> what's already been built.

>
> clone controllers were supposedly primary motivation for future system
> project ... an extrodinary complex machine with complex instructions.
> future system was targeted at completely replacing 360/370 and as
> different from 360 as 360 had been different from prior computer
> generations. future system was canceled w/o ever being announced
> ... some past posts
> http://www.garlic.com/~lynn/submain.html#futuresys
>
> Amdahl gave a talk in large MIT auditorium in the early 70s about
> leaving ibm and starting his own clone processor company. he was asked
> what justification he used with investors regarding his company. he
> replied that that ibm mainframe customers had already invested $200B in
> developing mainframe software ... and even if ibm were to completely
> walk away from 360 (which might be construed as veiled reference to
> future system project), that would be enough software to keep him in
> business through the end of the century.
>
> it has been claimed that the pre-occupation with future system (going to
> completely replacing 360/370) allowed the 370 product pipeline to go
> dray. with the demise of the future system effort, there was mad rush to
> get products back into the 370 hardware and software pipeline. However,
> the lack of 370 products is claimed to have contributed to allowing
> clone processors (like amdahl's) to gain a foothold in the market.
>
> I've also claimed that big motivation for John doing 801/risc was to go
> in the opposite extreme from what was going on with future system. lots
> of past posts mentioning 801, risc, iliad, romp, rios, power, power/pc,
> etc http://www.garlic.com/~lynn/subtopic.html#801
>
> this is old email mentioning the mip lisp machine group trying to get
> 801/risc chips
> http://www.garlic.com/~lynn/2003e.html#email790711
>
> for other drift ... this talks about shootout between QBE & system/r
> http://www.mcjones.org/System_R/SQL_Reunion_95/sqlr95-Shoot-ou.html
>
> original relational/sql was done on vm370 ... and compare&swap was part
> of the 370 instruction set. The discussion regarding compare&swap (in the
> above) strayed a bit (& got the details wrong).
>
> charlie had invented compare&swap instruction doing parallel, fine-grain
> multiprocessor locking work on cp67 (360/67 virtual machine precursor to
> vm370) at the science center ... misc. past posts mentioning the science
> center (note compare&swap name was chosen because CAS are charlie's
> initials)
> http://www.garlic.com/~lynn/subtopic.html#545tech
>
> initial attempts to get compare&swap into 370 architecture were
> rebuffed. the favorite son operating system claiming that test&set
> instruction was more than adequate for multiprocessor operation. the
> challenge was that to get compare&swap instruction into 370
> architecture, a non-multiprocessor specific use needed to be
> created. Thus was born the example use for application use ... still
> included in current principles of operation.
> http://publibz.boulder.ibm.com/cgi-bin/bookmgr_OS390/BOOKS/dz9zr003/A.6?DT=20040504121320
>
> since then many hardware architectures have implemented comapre&swap (or
> very similar instructions) and have been widely adapted by multithreaded
> applications (including most DBMS implementations) ... regardless of
> whether running in single processor or multiprocessor environment. misc
> past posts mentioning multiprocessor work and/or compare&swap
> instruction
> http://www.garlic.com/~lynn/subtopic.html#smp
>
> misc. past posts mentioning original relational/sql implementation
> http://www.garlic.com/~lynn/submain.html#systemr
>

Thanks, very interesting. Regarding C&S, I don't know that that has anything to do with database logic per se, rather it's for serialization in general. Regarding VM/CMS I used to know a guy who did a lot of work on the sql/ds optimizer. I remember him scoffing in the early 1990's as various improvements to the DB2 optimizer came out: "ha, sql/ds had that five years ago!"

I still imagine that Gene Amdahl had a different vision of complex instructions, ones that would be closer, in a sense, to applications. I knew another guy who remembered the Amdahl processors being planned on a whiteboard in Amdahl's office at IBM. But he lost control of his company before he even went into production when IBM announced 'virtual storage' he had to change his designs and needed Fujitsu money to keep going. Ten years ago, at a ripe old age, he was still working full days at one of his successor companies, if his secretary was at lunch, he'd answer the 'phones himself. Received on Fri Nov 13 2009 - 18:15:02 CST

Original text of this message