z/VM 50th

z/VM 50th

I had taken 2 credit hr intro to fortran/computers, at the end of semester was hired as student programmer and then within a year of the 2hr class, was hired fulltime responsible for os/360. Univ. had been sold a 360/67 for tss/360, but never came to production fruition and so ran w/os360 as 360/65. Univ. shutdown datacenter on weekends and I had the whole place to myself (although 48hrs w/o sleep made monday classes hard). Last week of Jan1968, 3 people came out to univ to install cp67 (3rd installation after CSC and MIT lincoln labs), I mostly played with it in my weekend time ... initially rewritting lots of pathlengths to improve OS/360 throughput. I started with os/360 MFT jobstream that took 322sec, but 856sec under CP67 (CP67 CPU 534sec). After a few months, I had it down to 435secs (CP67 CPU 113sec, down from 534sec). Part of SHARE presentation in old archived post

https://www.garlic.com/~lynn/94.html#18

I was included in the IBM CP67 announce at the spring 68, Houston SHARE meeting (54 years ago).

I then do a lot of work on scheduling (dynamic adaptive resource management), I/O throughput (from fifo single request to ordered seek queuing, chained page i/o request, improving 2301 from 80/sec to 270/sec, page replacement (from not-in-queue then random to reference bits). Before I graduate, I'm hired into small group in the Boeing CFO office to help with the formation of Boeing Computing Services (consolidate all dataprocessing into independent business unit to better monetize the investment). I thought Renton datacenter possibly largest in the world, couple hundred million in 360s, 360/65s arriving faster than they could be installed, boxes constantly staged in hallways around machine room. Lots of politics between manager of Renton datacenter and CFO (who only had 360/30 for payroll up at Boeing field ... although they enlarge the machine room for 360/67 for me to play with when I'm not doing other stuff).

When I graduate, I leave Boeing and join the IBM science center. One of my hobbies after joining IBM was enhanced production operating systems for internal datacenters.?After decision to make all 370s virtual memory, several people split off from the science center (on the 4th flr) and took over the IBM Boston Programming center on the 3rd floor for VM/370 development. When the VM370 group outgrew the 3rd floor (they only had part, the rest was a three letter gov. agency), they moved out to the empty IBM SBC bldg at Burlington Mall. The initial morph of CP67->VM370 dropped or significantly simplified lots of stuff from CP67 (including dropping multiprocessor support and lots of stuff I had done as undergraduate) ... for VM370 Release 1 (50yrs ago).

I start CP67->VM370 migration with release 2. We had a automated benchmarking system that could vary configuration, workloads, parameters, etc. However, initially VM370 would crash every time ... so the first things that I had to migrate were the CP67 kernel serialization and integrity features in order to complete benchmarks w/o crashing. Some old email getting VM370 Release 2 to production quality for shipping as internal CSC/VM.

https://www.garlic.com/~lynn/2006w.html#email750102

https://www.garlic.com/~lynn/2006w.html#email750430

Much of this was during the IBM Future System period which was going to completely replace 370 and internal politics were killing and/or suspending lots of 370 activity (claim is that the lack of new 370 during FS contributed to clone 370 makers getting market foothold). Then with the death of FS, there was mad rush to get stuff back into the 370 product pipeline, including kicking off the quick&dirty 3033&3081 efforts in parallel. Some refs:

https://www.jfsowa.com/computer/memo125.htm

https://people.cs.clemson.edu/~mark/fs.html

Some of the mad rush after FS implodes, contributed to picking up a little bit of my CSC/VM for release 3 (a few of the integrity fixes and DCSS was small subset of my CMS page-mapped filesystem and shared segment enhancements). Then the head of POK managed to convince corporate to kill VM370, shutdown the development group and transfer all the people to POK (or otherwise MVS/XA wouldn't ship on time). They weren't planning on telling the group to the very last minute to minimize people that might escape the move. It managed to leak and several people escaped (this was during early DEC VAX days and joke was that the head of POK was major contributor to DEC VMS). There was witchhunt for the source of the leak, fortunately nobody gave me up. Endicott managed to save the VM370 product mission but had to reconstitute a development from scratch.

Endicott also cons me into doing analysis for ECPS (moving VM370 kernel 370 instructions into microcode) ... initially for virgil/tully (138/148). Was to find the highest executed 6k bytes of the kernel for moving into the available space for microcode. Old post with the initial analysis (6k bytes accounted for 80% of kernel execution and would move into microcode at 10:1 speedup).

https://www.garlic.com/~lynn/94.html#21

I was then con'ed into running around the world helping Endicott present the 138/148 & ECPS business cases to country business planners.

trivia: one of my longest and biggest internal customers was the world-wide sales&market support online HONE system (dating back to CP67 days). In the mid70s, the US HONE datacenters were consolidated in Palo Alto (trivia: when Facebook 1st moves into silicon valley, it is into a new bldg built next door to the old HONE datacenter), enhancing VM370 with single-system image, cluster, loosely-coupled having load-balancing and fallover support. I then add multiprocessor support to a release 3 CSC/VM ... initially so HONE can add a 2nd processor to each system ... for 16 processor operation. This century I joked about IBM not releasing software before its time, z/VM announced single-system-image support in 2009 (more than 30yrs after HONE was running it).

*** 370 virtual memory footnote: A decade ago, I was asked to track down the decision to make all 370s virtual memory. I found somebody that reported to the executive making the decision. Basically MVT storage management was so bad, regions had to be four times larger than typically used. As a result, a standard 1mbyte 370/165 only had four regions, insufficient to keep the machine busy and justified. Going to virtual memory (initially VS2/SVS, very similar to running MVT in CP67 16mbyte virtual machine), would allow increasing the number of regions by a factor of four times with little or no paging. Ludlow is doing initial work on MVT for SVS on 360/67 and biggest code is for EXCP/SVC0 making copy of passed channel programs, replacing virtual addresses with real ... and crafts in CP67 CCWTRANS (that provides similar function for virtual machine channel programs) into EXCP/SVC0. Old archived post with pieces of the exchanged email:

https://www.garlic.com/~lynn/2011d.html#73