On Wednesday, 1994-07-27  Donald Macmillan wrote to David Noon about
"DA/2 Distributed Applica" as follows:

DM> our comments re FORTRAN are interesting. I discounted the FORTRAN
DM> code as a  source of the problem, since the code was identical in
DM> the  DOS and OS/2 compiles (ie the source code was identical).  My
DM> rationale was that Watcom 9.5 compiler would produce  code which was
DM> more or less equally optimized and would  have approximatelt the
DM> same run-time. Are you implying that  the FORTRAN code would be the
DM> culprit, ie. may be running  30% slower as a DLL under OS/2? Or
DM> worse still, maybe the  FORTRAN code is running 100% slower and any
DM> improvement  from OS/2 is offset by retarded FORTRAN?

Hi Donald,

This is precisely what I am saying. If you do not check your FORTRAN
code for memory accesses that are well behaved in a virtual storage
environment you run the risk of excessive page faults. The magnitude of
the performance hit will vary with the application, so I could not
offer a percentage.

The reason FORTRAN is the most frequent culprit is because it stores
its arrays "at right angles" to most other languages, all of which are
more suited to implementing numerical algorithms as specified in most
journals; you need to rotate the algorithm 90 degrees to suit FORTRAN.

Regards

Dave

___
 X KWQ/2 1.2e X Computing ... where senility is your only relief.

--- Maximus/2 2.01wb