Dave Kopischke wrote:
On Tue, 4 Jul 2006 10:49:45 +0100, Jim McAlpine wrote:
I'm looking at the Enterprise COBOL Performance Tuning paper regarding the
ARITH(EXTEND) compiler option which says -
.....
| On the average, ARITH(EXTEND) was 1% slower than ARITH(COMPAT),
with a range of equivalent to 38% slower.
We're enhancing some processes and converting to ARITH(EXTEND) as a
result of increased field lengths. Initial testing is showing about a 30%
increase in CPU time.
Has anyone else done this conversion and experienced similar performance
degradation ????
I expect a certain performance hit, but this seems high to me. I also note this
old post on the same subject. Where can I find a copy of this "COBOL
Performance Tuning" document ???
Thanks again,
Dave K.
When you compile with ARITH(EXTEND), _all_ computations are done
in "extended mode" (carried to extra places of precision)
Intrinsic functions that return floating point numbers are also
modified to return extended precision floating point numbers
So of course, the answer to your question is, "it depends"; on
what your programs are doing. Still, 30% seems high? If you are
not running on a machine that supports the new floating point
registers and instructions you may also be seeing some software
simulation going on, don't know.
Kind regards,
-Steve Comstock
The Trainer's Friend, Inc.
303-393-8716
http://www.trainersfriend.com
-- to be included in our opt-in list of announcements of
-- new courses and other products and services from The
-- Trainer's Friend, send an email to [EMAIL PROTECTED]
----------------------------------------------------------------------
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
