I was running an old script to refresh my memory of how the halo mass function is used in YT, and it just siphoned all the memory out of my laptop on a tiny dataset. As I recalled I was able to test the halo mass function extensively on my laptop with this 64 cube unigrid run. Thinking memory is the issue (mine or my laptop's), I then tried to run the script on triton with the TCC node, it still hangs and runs out of memory, I then tried the PDAF node and got the same results even after updating to the latest dev (22ac529a1dab (yt) tip). link to the script: http://paste.enzotools.org/show/1815/ Or is there something obviously wrong with the script? From G.S.
Hi Geoffrey,
Thinking memory is the issue (mine or my laptop's), I then tried to run the script on triton with the TCC node, it still hangs and runs out of memory, I then tried the PDAF node and got the same results even after updating to the latest dev (22ac529a1dab (yt) tip).
I see the same issue as you. I turned down the precision of the integrator in the halo mass function which should make these kind of memory overruns less likely. Please update your install and let me know if A) you still see problems or B) if the fit looks fishy (due to the lowering of the integrator's precision). Good luck! -- Stephen Skory s@skory.us http://stephenskory.com/ 510.621.3687 (google voice)
Hi Stephen,
I've been comparing results of the halo mass function with the old ones on
the small dataset, it seems the new fit function is systematically higher
than the old ones. I've also ran it on a 20 Mpc 800 cube unigrid, and the
fit is also higher than the simulated data by almost 4x. I'll send the
pictures in another private email so I don't spam the list.
Just want to let people know it no longer hangs due to not enough memory
during calculation of the HMF. Does anyone have a halo profile data at hand
they can just check the halo mass function against the Warren fit in
particular? Calculating the halo mass function is quick if the haloes are
already profiled and saved, took less than a minute on a single processor
for the 800 cube.
From
G.S.
On Wed, Sep 21, 2011 at 8:00 PM, Stephen Skory wrote:
Hi Geoffrey,
Thinking memory is the issue (mine or my laptop's), I then tried to run the script on triton with the TCC node, it still hangs and runs out of memory, I then tried the PDAF node and got the same results even after updating to the latest dev (22ac529a1dab (yt) tip).
I see the same issue as you. I turned down the precision of the integrator in the halo mass function which should make these kind of memory overruns less likely. Please update your install and let me know if A) you still see problems or B) if the fit looks fishy (due to the lowering of the integrator's precision).
Good luck!
-- Stephen Skory s@skory.us http://stephenskory.com/ 510.621.3687 (google voice) _______________________________________________ Yt-dev mailing list Yt-dev@lists.spacepope.org http://lists.spacepope.org/listinfo.cgi/yt-dev-spacepope.org
participants (2)
-
Geoffrey So
-
Stephen Skory