[Radiance-general] No more Mr. Nice Guy

[Reinhart, Christoph]

Dear all,

Greg's email is not only relevant for Mac users but also for the
friendly but frequently overlooked WINDOWS crowd:)

For the Daysim Windows version, I am compiling a number of Radiance
programs. Specifically, I am calling rtrace via a C-Program called
gen_dc to calculate a whole set of daylight coefficients. In case rtrace
has been compiled with NICE=, it runs with a "low" priority whereas
without the NICE=, it runs with "normal" priority. In the latter case,
the simulation times can become substantial as just about any other job
under Windows runs with "normal" priority. In case you are using
Radiance under Windows (cygwin etc.), this might be worthwhile looking
into as well. One of my students reported that the Lightstudio Radiance
binaries run significantly faster than Desktop Radiance (but this may be
the case for numerous other reasons as well).

Christoph

-----Original Message-----
From: radiance-general-bounces at radiance-online.org
[mailto:radiance-general-bounces at radiance-online.org] On Behalf Of Greg
Ward
Sent: Thursday, December 15, 2005 11:51 PM
To: Radiance general discussion
Subject: [Radiance-general] No more Mr. Nice Guy

Hi All,

This message is most relevant to Mac users, but introduces a change that
may affect rpict and rtrace performance in future releases on all
platforms.  Specifically, I have taken out the NICE= macro settings from
the Rmakefile in src/rt, so future builds will run all Radiance programs
at the standard user priority.

I made a fortuitous discovery whilst playing around with the very cool
(no pun intended) TemperatureMonitor freeware <http://
www.bresink.com/osx/TemperatureMonitor.html> on my new PowerMac G5 quad.
It's lots of fun to watch the processor temps go up and down with load.
The funny thing I noticed was that changing the Energy Savings settings
had an unexpected effect on temperatures, which are in turn related to
power consumption and (presumably) processor performance.

The Energy Saver options lists three processor performance settings,
"Highest," "Automatic," and "Reduced."  With the machine idle, there
were only slight differences between the three; I had expected to see
more, especially on the "Highest" setting...  The biggest difference was
between "Highest" and "Reduced" with all four processors busy, and that
was no surprise.  What was a surprise was that when I had four rpict
jobs going in parallel, there was a big difference between  
the temperatures using the "Highest" versus "Automatic" settings.   
Apple documentation states that the performance penalty for using
"Automatic" should be very small, so I figured the temperature change
would also be small.  That's when I decided to measure actual
performance.

What a difference!  Between the "Highest" and "Automatic" settings, I
noticed a performance difference of over 60%, where I expected to see
little or none.  What I did expect was a performance hit moving to the
"Reduced" setting, which indeed there was, but it was only  
another 13% worse than "Automatic."  This really seemed wrong to me.   
That's when I decided to start playing around with the "nice" level of
rpict.

The "nice" setting on a process is a way to tell Unix that a job is not
a top priority, and interactive processes should be given more of the
CPU if there are any running.  However, in the past, it hasn't much
affected the performance of a process if there are no other jobs to
contend with.  I put nice() calls in Radiance way back when multi- user
Unix's were the rule, and we had to peacefully coexist with a lot of
other people who might get annoyed if I hogged all the system resources,
as Radiance is wont to do.

In short, I found out that removing the nice() system call from rpict
achieved the "Highest" performance figure even when the Energy Saver
processor speed was set to "Automatic."  (The "Reduced" performance was
not affected.)  Apparently, Mac OS X 10.4 (at least) uses the process
"nice" setting to decide whether or not to kick the CPU speed up a
notch.  This might be a new feature of the OS, as I don't have access to
a maching running Jaguar or Panther to try it there.

The bottom line in all this is that if you own a Mac and you're running
Radiance on it, be sure to use the "Highest" setting in the Energy Saver
control panel, because if you don't, your rendering might just be
running 60% slower than you would like it to.  Option B is to remove the
-DNICE= arguments from the build lines in src/rt/ Rmakefile and remove
the affected object files (rpmain.o and
rtmain.o) then rerun "rmake install" in the rt directory.  Or, download
Saturday's HEAD release and build that.  (I think I'm too late for
tomorrow.)  Then, you can leave the processor performance on
"Automatic," which is the default for normal operation.

I wish I had made this discovery a couple of years ago....

-Greg


_______________________________________________
Radiance-general mailing list
Radiance-general at radiance-online.org
http://www.radiance-online.org/mailman/listinfo/radiance-general