[R-sig-hpc] parallel version of the bootstrapping boot function

[Laurence Baldwin]

Hi,
I am busy developing a parallel version of the boot function from the
boot package, at present it is fully functional and passes all the
RUnit tests I have written for it. The parallel version named "pboot"
was developed using the SPRINT framework (http://www.r-sprint.org/)
for communication.

The speedup has been benchmarked using the 4 different bootstrapping
examples from the boot manual
(http://stat.ethz.ch/R-manual/R-devel/library/boot/html/boot.html)
with the number of replications increased to give longer run times.
These results show decent speedup depending on the problem. However I
do not think these examples are typical of large bootstrapping
problems with long execution times (Hopefully bigger problems will
give better speedup).

Before I look at improving the speedup and refactoring the code,  I
would like to get some real world examples of bootstrapping problems
that people are trying to speedup. This would also help me test if my
version is 100% compatible and bug free.

$ mpiexec -n 6 R --slave -f pboot-Speedup-Benchmark.R
             	boot  		pboot		  speedup
weight     42.140 		24.392	 	1.727616
freq       	43.065	 	8.401		 5.126175
ordinary 	  41.654 	 	7.502 		5.552386
parametric 36.993	12.975 		2.851098

Further benchmarks are on the project site
http://code.google.com/p/baldy/ along with source code in SVN
http://code.google.com/p/baldy/source/checkout. All the code is in
beta but passes all the RUnit tests I have written so far and the
results can easily be verified against the standard version of boot.

If you are using the boot function and would like to speed it up on
HPC or SMP systems please contact me with examples.

Regards
 Laurence