[R-sig-Geo] GStat 'predict.c' parallel estimation
Tim Peterson
timjp at unimelb.edu.au
Thu Aug 27 02:45:17 CEST 2015
Hi all,
I'm considering editing the GStat kriging file 'predict.c' (while loop
at line 151-165) for parallel estimation. I require parallel estimation
for both kriging and indicator simulation and will be applying it to a
region of ~30 million grid cells, which will be repeated for 300 time
points.
I appreciate that snow can be used for parallel kriging
(https://r-forge.r-project.org/scm/viewvc.php/pkg/demo/snow.R?view=markup&revision=89&root=gstat)
but, because of my computation burden, I require parallel calculation
using Xeon Phi cards - hence I think the changes should be in
'predict.c'. Also, in looking at the GStat C-code I suspect that the
global variables in 'glvars.c' may complicate the parallel calculations.
So anyway, before I embark upon this task (most likely using openMP),
I'd be grateful if anyone could share tips or thoughts (and my edits
will be freely shared.)
Also, some may be questioning the parallel estimation of each
simulation. Considering that I'm estimating ~30 million cells, and will
only be using ~70 cores per simulation, I am happy to accept the
deficiency that recently estimated cells will not be used in the
estimation of nearby cells within a concurrent parallel cycle. However,
the estimates do need to be accessible to latter parallel cycles.
Lastly, my backup option is to use openMP to parallelise the GSLib
Fortran subroutines 'cokb3d.for' and 'sisim.for' (ie outside of R). This
is mainly because the subroutines are fairly standalone and don't use
pointers - and hence possibly less problematic to make parallel.
Many thanks,
Tim
----------------------
Dr. Tim Peterson
The Department of Infrastructure Engineering
The University of Melbourne, 3010 Australia
T: +61 3 8344 9950, M: +61 0438 385 937
Dept. profile :
http://www.ie.unimelb.edu.au/people/staff.php?person_ID=141135
Research Gate : https://www.researchgate.net/profile/Tim_Peterson7
Google Scholar:
http://scholar.google.com.au/citations?user=kkYJLF4AAAAJ&hl=en&oi=ao
More information about the R-sig-Geo
mailing list