Hi All,
I’ve been testing the -n option of rtrace, but I’m not getting speed
increases that make sense. My test job takes rtrace ~30 minutes on one
core. With four cores I only get down to 24 minutes (and ~25 minutes
for 2 and 3 cores). I can see in system monitor that it is using the
right number of cores. I’m not expecting exactly 1/4th the run time,
but only 20% faster doesn’t sound quite right. Any thoughts?
Thanks,
Cramer
···
–
**Cramer Silkworth
212 219 2255 | [email protected]
Transsolar Inc. | 134 Spring Street Suite 601 | New York, NY 10012
Transsolar Climate Engineering
Technical consulting for energy efficiency and environmental quality in
buildings.
New York - Stuttgart - Munich**
Hi Cramer,
Can you tell us the command line you are using to run your rtrace job?
-Jack
···
--
# Jack de Valpine
# president
#
# visarc incorporated
# http://www.visarc.com
#
# channeling technology for superior design and construction
Cramer Silkworth wrote:
Hi All,
I've been testing the -n option of rtrace, but I'm not getting speed increases that make sense. My test job takes rtrace ~30 minutes on one core. With four cores I only get down to 24 minutes (and ~25 minutes for 2 and 3 cores). I can see in system monitor that it is using the right number of cores. I'm not expecting exactly 1/4th the run time, but only 20% faster doesn't sound quite right. Any thoughts?
Thanks,
Cramer
--
*Cramer Silkworth
212 219 2255 | [email protected] <mailto:[email protected]>
Transsolar Inc. | 134 Spring Street Suite 601 | New York, NY 10012
Transsolar Climate Engineering
Technical consulting for energy efficiency and environmental quality in buildings.
New York - Stuttgart - Munich*
------------------------------------------------------------------------
_______________________________________________
Radiance-general mailing list
[email protected]
http://www.radiance-online.org/mailman/listinfo/radiance-general
Jack was right to ask for more info, and your commands will be helpful, but
in the meantime this may be useful (in particular, skip to the bottom of the
post where Andy McNeill contributes the useful part):
http://www.rumblestrip.org/interests/light/rtrace-multiprocessing-option-ini
tial-test-results/
- Rob
···
On 6/24/10 2:33 PM, "Cramer Silkworth" <[email protected]> wrote:
Hi All,
I've been testing the -n option of rtrace, but I'm not getting speed increases
that make sense. My test job takes rtrace ~30 minutes on one core. With four
cores I only get down to 24 minutes (and ~25 minutes for 2 and 3 cores). I can
see in system monitor that it is using the right number of cores. I'm not
expecting exactly 1/4th the run time, but only 20% faster doesn't sound quite
right. Any thoughts?
Thanks,
Cramer
Hi Rob,
Shows how good my memory is. That is a great test that you did along with the final eye opening clarification from Andy.
-Jack
···
--
# Jack de Valpine
# president
#
# visarc incorporated
# http://www.visarc.com
#
# channeling technology for superior design and construction
Guglielmetti, Robert wrote:
Jack was right to ask for more info, and your commands will be helpful, but
in the meantime this may be useful (in particular, skip to the bottom of the
post where Andy McNeill contributes the useful part):
http://www.rumblestrip.org/interests/light/rtrace-multiprocessing-option-ini
tial-test-results/
- Rob
On 6/24/10 2:33 PM, "Cramer Silkworth" <[email protected]> wrote:
Hi All,
I've been testing the -n option of rtrace, but I'm not getting speed increases
that make sense. My test job takes rtrace ~30 minutes on one core. With four
cores I only get down to 24 minutes (and ~25 minutes for 2 and 3 cores). I can
see in system monitor that it is using the right number of cores. I'm not
expecting exactly 1/4th the run time, but only 20% faster doesn't sound quite
right. Any thoughts?
Thanks,
Cramer
_______________________________________________
Radiance-general mailing list
[email protected]
http://www.radiance-online.org/mailman/listinfo/radiance-general
Jack, Rob, et
al,
I read Rob’s posting and have been playing with -aa, but I’m not sure I
know what I’m doing. Here’s what I’ve got:
this: rtrace -n 4 -I+ -h -ab 8 -ar 128 -aa 0.1 -ad 1024 -as 512 -e
grid.log cie10k.oct < grid.pkt > grid.out
takes 23.5 minutes, for 3780 measurement points.
this: rtrace -n 4 -I+ -h -ab 8 -ar 128
-aa 0 -ad 1024 -as 512 -e grid.log cie10k.oct < grid.pkt >
grid.out
(-aa 0 is the only change) takes 13 seconds for the same measurement
points, but yields daylight factor results (calc’d with an rlam|rcalc
line following this rtrace call) 30% lower on average.
Andy’s email to Rob via this list (and posted as an update to Rob’s
blog post) mentions an ambient file. Is there a way to “pre-generate”
this file before starting rtrace, such that each process will use it?
Or am I just totally confused here (=yes).
Thanks,
Cramer
···
–
**Cramer Silkworth
212 219 2255 | [email protected]
Transsolar Inc. | 134 Spring Street Suite 601 | New York, NY 10012
Transsolar Climate Engineering
Technical consulting for energy efficiency and environmental quality in
buildings.
New York - Stuttgart - Munich**
Hi Cramer,
You don't need to "pregenerate" the ambient file for rtrace purposes, but you can. The disadvantage is that it will take about as long to pregenerate the file as it does to run rtrace, though the subsequent run should go very fast.
Sharing values via an ambient file with -aa non-zero allows parallel processing to actually work in Radiance. It doesn't require a precalculation, though. The only reason to do that is to get slight more accurate and/or smoother results.
Cheers,
-Greg
···
From: Cramer Silkworth <[email protected]>
Date: July 2, 2010 12:48:40 PM PDT
Jack, Rob, et al,
I read Rob's posting and have been playing with -aa, but I'm not sure I know what I'm doing. Here's what I've got:
this: rtrace -n 4 -I+ -h -ab 8 -ar 128 -aa 0.1 -ad 1024 -as 512 -e grid.log cie10k.oct < grid.pkt > grid.out
takes 23.5 minutes, for 3780 measurement points.
this: rtrace -n 4 -I+ -h -ab 8 -ar 128 -aa 0 -ad 1024 -as 512 -e grid.log cie10k.oct < grid.pkt > grid.out
(-aa 0 is the only change) takes 13 seconds for the same measurement points, but yields daylight factor results (calc'd with an rlam|rcalc line following this rtrace call) 30% lower on average.
Andy's email to Rob via this list (and posted as an update to Rob's blog post) mentions an ambient file. Is there a way to "pre-generate" this file before starting rtrace, such that each process will use it? Or am I just totally confused here (=yes).
Thanks,
Cramer
--
Cramer Silkworth
212 219 2255 | [email protected]
Transsolar Inc. | 134 Spring Street Suite 601 | New York, NY 10012
Transsolar Climate Engineering
Technical consulting for energy efficiency and environmental quality in buildings.
New York - Stuttgart - Munich
What happens if you keep -aa set to 0.1 and add the option -af grid.amb ?
···
____________________________________________________________
Electronic mail messages entering and leaving Arup business
systems are scanned for acceptability of content and viruses