Summary - IBM 1130 User Manual

Computing system
Hide thumbs Also See for 1130:
Table of Contents

Advertisement

Summary
To recapitulate the lessons learned in the preceding
three case studies, performance depends on five
major factors:
1. The size of the program. When writing any
program, you should anticipate problems with
core storage and performance. Plan pro-
grams of reasonable scope, and code them as
a series of LINKs, if at all possible.
2.
The subroutines required by the program.
Realize that many seemingly innocent
FORTRAN statements can cause sizable sub-
routines to be included in your core load.
Some examples are PAUSE, STOP, FlND,
division, use of the data switches, etc.
FORTRAN control cards can have a similar
effect -- for example, unnecessary *IOCS
cards, the TRAC E, etc.
3.
The way the program is structured. When
flowcharting and coding your programs,
always keep in mind the location of the disk
arm, so that you do not invite excessive arm
movement between the overlay area and the
data area. Place as little coding as possible
between disk READ/WRITE loops so that the
chance of an intervening overlay is reduced.
Figure 90.11 shows the various combinations
of data files and overlays.
Note that the location of the overlay has a
great effect on performance. If you must
move the disk arm from one area to the other,
you can at least try to minimize the number
of times it is required (or reduce the distance
involved, by making data files compact).
4. The overlay scheme used. If your program
is of such magnitude that some overlaying is
required, you should have a good feel for
how each works and how each can affect per-
formance. Figure 90.11 shows that there is
no differentiation made between LOCALs,
SOCALs, and LINKs -- they are all overlays.
Note also that the number of times an over-
lay is required is not as important as the disk
Section
Subsections
Page
90
40
I
40
01
arm movement that may be necessary to get
it. For this reason you should take particular
care
to
avoid causing an overlay to be placed
in between disk READ/WRITE statements.
LOCALs, because they are selected by
the programmer, will often yield better per-
formance than SOCALs, which are chosen by
the CLB according to predetermined rules.
However, if you select LOCALs without
regard to their effect on performance, it is
possible that they can slow down execution
time even more than SOCALs.
5. The size and location of the data file. Since
you are concerned with minimizing disk arm
movement time, you should try to shorten
the distance involved.
The overlay area is always at the end of the
UA or at the beginning of WS, whichever way you
prefer to look at it. The data files may be either:
In the UA or FX, if you have put them
there with the *STOREDATA card
• At the end of UA (beginning of WS), if you
have not used a *STOREDA TA or *FILES
card
If you have a temporary file, in WS, your arm
movement times will be minimized, since the
files and the overlays are as close as they can
be. If your file is in the UA, however, the
picture may be quite different, depending on
how "deep"
it
lies
in
the UA. If a DUMPLET
shows that there is a great deal of distance
between the file and the end of UA, you should
consider moving the file. Figure 90.12 shows
three possible situations.
The key to gaining good program performance is
knowledge:
• Knowledge of the way
in
which the three
overlays work
• Knowledge of the basic workflow of your
program

Advertisement

Table of Contents
loading

Table of Contents