SUMMARY - System time vs. User time

From: Mahesh Subramanya (
Date: Thu Mar 21 1991 - 20:22:56 CST

My problem was that a particular set of FORTRAN programs on our
system had system times vs user time ratios of as much as 20:1. The
programs essentially dealt with relatively large arrays being filled
with values using calculations which were quite CPU intensive.

Interestingly, I received around 15 "me-toos", a couple of requests
to summarize, and three suggestions. They were

>From: Tad Guy <>
>Sounds like you're being bitten by the bad swapping alg in SunOS 4.
>We've disabled it here (with the patch below) and have had

>significant performance improvements.
>root# adb -w /vmunix

>root# reboot

I had speculated that the problem might be due to the PMEG problem,
or the above. However, profiling showed that the problem was neither
of the abopve, but in fact, was due to

>From: (Hal Stern - Consultant)
>From: sundev!fletch! (Kevin Sheehan {Consulting Poster
>when your fortran program is done, does it print messages about
>detected underflow conditions? the larger amounts of system
>time generally indicate a problem with gradual underflow, instead
>of "abrupt underflow", which causes the floating point unit to
>"round down" small absolute values to zero.

>with your fortran compiler, you should have received a copy of
>the "numerical computation guide". the solutions to this problem
>are discussed in the section on ieee exception handling
>going to single precision makes this worse --> this implies that
>it's really an underflow problem, since the floating point unit
>then has to spend *even more* effort to maintain "precision" in
>the presence of underflow. fewer bits of precision == more work.
>i doubt this is swapping/MMU related. larger quantities of system
>time indicate an exception handling problem. if you were really
>bound by swap or page replacement, you'd see large quantities of
>*idle* time as the system waited for pages to come off of the disk.

This indeed was the problem.. Thanx very much y'all....

Mahesh Subramanya INTERNET:
Senior Analyst

Office of University Computing

University of Notre Dame Voice: (219) 239-5600 x6421
Notre Dame, IN 46556

This archive was generated by hypermail 2.1.2 : Fri Sep 28 2001 - 23:06:12 CDT