I very largely agree with Gerhard Postpischl's comments.
The question of the effects of new machine instructions on execution times
does, however, require further comment.
The first, surpassingly obvious thing that must be said about new instructions
is that not every mainframe currently in use is equipped with them. In
consequence an ISV product or an IBM translator like the HLASM is likely to
avoid use of new instruction N until it is judgesd that all of the machines its
customer base uses are equipped with that new instruction, which can take a
preternaturally long time.
The second thing that must be said is that IBM translators, those for z/OS
anyway, are mostly equipped with an ARCH[itectural level] option which
specifies which instruction set is to be used in compiled code. If the ARCH
level for a compilation is not set high enough to include new instructions they
go unused in compiled code. (An obvious corollary is that if, say, a C or PL/I
application is not recompiled by a compiler version that supports a new, higher
architectural level, the availability of that level on some target machine will
have no effect upon the execution times of that application on it.)
Now I have conducted no proper survey using a probability sample of mainframe
shops, but it is my impression that---excepting a few leading-edge shops from
these strictures---most of them give too little attention to 1) specifying
appropriate ARCH levels for compilations, 2) recompiling applications to permit
the code compiled for them to reflect the availability of new instructions, or
even 3) making new versions of compilers that support higher ARCH levels
available in timely fashion.
The availability of new instructions does come over time to be reflected in the
code skeletons used to compile applications, but this happens much more slowly
than it should. Inertia and lethargy must be and are paid for. At best
obsolescent techniques are often used long after they could have and should
have been replaced.
Moreover, concern with these 'bits and bytes' issues has never been great on
the applications sides of most mainframe shops; and there is every indication
that it is declining as even minimal understanding of them disappears.
John Gilmore Ashland, MA 01721-1817 USA
----------------------------------------------------------------------
For IBM-MAIN subscribe / signoff / archive access instructions,
send email to [email protected] with the message: GET IBM-MAIN INFO
Search the archives at http://bama.ua.edu/archives/ibm-main.html
