🚀 go-pugleaf

In a recent article mnyk...@cc.helsinki.fi writes:
>An acquintance of mine was hired to speed up a COBOL accounting program
>which used some 8 hours to process some 1000's of transactions. She
>discovered that the program processed the first transaction, then
>REWOUND the file, sought the second, bypassing the first, rewound,
>sought the third, etc.. Instead of some 1000's of file operations the
>program performed some 1000000's!

Sounds like a program I had the displeasure of running on my system in
the late 1960's.  It was called SEMIS (State Extension Management Information
System), and as I recall it was distributed by USDA to the various states 
to keep track of the agricultural extension service activity.  The problem
was that it was written in COBOL, but the design criteria called for it to
be usable on just about any platform with a COBOL compiler, including those
which supported only the primitive COBOL subset.  Since the subset did not
include direct access I/O, it simulated the facility by REWIND/READ/READ...READ
which didn't do much for speed.  On an IBM 360/40 (which was not known for
its CPU speed) we recorded a wallclock-to-CPU time ratio of 3600:1. 

Unfortunately, the university had committed to run the program for the ag
campus without bothering to talk to the computer center, so we couldn't
get rid of it until the agreement expired.  Fortunately someone apparently
realized what it was doing to the (rather expensive) computer center
service levels and didn't renew the agreement.