[email protected] (Anne & Lynn Wheeler) writes:
> however, major heavy weight (besides avoiding TCB tasking) was
> OPEN/CLOSE for task. CICS did batch open at startup ... disk accesses
> and pathlength for OPEN/CLOSE would have swamped typical task disk
> accesses and execution time. First bug I shot was OPEN ...
> implementation stuffed some bits in DCB fields for specific BDAM
> options.  Library was using different set of BDAM options, the OPEN
> would fail and couldn't get it started. I had to zap some instructions
> to stop the DCB field fiddling in the CICS code.

re:
http://www.garlic.com/~lynn/2012h.html#78 Familiar
http://www.garlic.com/~lynn/2012i.html#7 Familiar

one of the things i started doing around release 11 was totally taking
apart the stage2 output from stage1 sysgen and reordering it so the
execution sequence would carefully place files and pds members on disk
for optimal disk arm seeking. for the univ. student workload, this
gained approx. three times increase in throughput. one of the major wins
was ordering the multitude of svclib open/close pds members

long ago and far away, part of presentation I made at
fall '68 SHARE in Atlanta
http://www.garlic.com/~lynn/94.html#18

That spring and summer, I had significantly rewritten sections of cp67
code (in this benchmark with mft14, reduced cp67 processing time from
534 cpu seconds to 113 cpu seconds).

CP67 never did quite make it to production at the univ., I just got to
play with it on weekends. Mostly the 360/67 ran in 360/65 mode with
os/360. One of the aggravations with carefully PDS member ordering was
PTFs replacing PDS members and destroying the careful ordering. Lots of
PTF activity over time could cut throughput in half ... and if a new
release build wasn't imminent, I would have to rebuild the current
release to get throughput back.

in any case, even with careful ordering ... open/close process still
involved fetching a large number of PDS members ... doing on a
transaction basis would have totally destroyed CICS thruput.

-- 
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