R help - Aug 2012 - parallel SNOW slower than single core?

Dear All,

I am learning parallel in R and start with the package "snow". I did a
test
about running time and the parallel version is much slower than the regulat
code. My laptop is X200s with dual core intel L9400 cpu.
Should I make more clusters than 2? Or how to improve the performance?

# install.packages("snow")
library(snow)
cl <- makeCluster(2)
t1 <- proc.time()
a <- c()
for (i in 1:1000)
{
a[i] <- sum(parSapply(cl, 1:15, get("+"), 2))
}
proc.time()-t1
t2 <- proc.time()
a <- c()
for (i in 1:1000)
{
a[i] <- sum(sapply(1:15, "+", 2))
}
proc.time()-t2

	[[alternative HTML version deleted]]

I'm not surprised -- you're doing a fairly trivial calculation so the
overhead of setting up the parallelization is likely more than the
benefit from halving the computation time. Do it on a larger
calculation and I imagine the performance will be better -- I might
just be pulling this out of thin air, but I think I remember reading
slides from Luke Tierney saying one had to be adding vectors of
size
>= 30k or so for parallelization to have a meaningful benefit.
Obviously sizes will be lower for operations more expensive than
addition.

Michael

On Thu, Aug 2, 2012 at 1:13 PM, Jie <jimmycloud at gmail.com>
wrote:
> Dear All,
>
> I am learning parallel in R and start with the package "snow". I
did a test
> about running time and the parallel version is much slower than the regulat
> code. My laptop is X200s with dual core intel L9400 cpu.
> Should I make more clusters than 2? Or how to improve the performance?
>
> # install.packages("snow")
> library(snow)
> cl <- makeCluster(2)
> t1 <- proc.time()
> a <- c()
> for (i in 1:1000)
> {
> a[i] <- sum(parSapply(cl, 1:15, get("+"), 2))
> }
> proc.time()-t1
> t2 <- proc.time()
> a <- c()
> for (i in 1:1000)
> {
> a[i] <- sum(sapply(1:15, "+", 2))
> }
> proc.time()-t2
>
>         [[alternative HTML version deleted]]
>
> ______________________________________________
> R-help at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide
http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.

On 02/08/2012 19:13, Jie wrote:
> Dear All,
>
> I am learning parallel in R and start with the package "snow". I
did a test
> about running time and the parallel version is much slower than the regulat
> code. My laptop is X200s with dual core intel L9400 cpu.
The OS matters far more.  Overheads on Windows are much higher than on 
Linux for example.
> Should I make more clusters than 2? Or how to improve the performance?
Not use such small tasks.  This is discussed in the 'parallel' vignette 
(and we do suggest you start with 'parallel' these days).
> # install.packages("snow")
> library(snow)
> cl <- makeCluster(2)
> t1 <- proc.time()
> a <- c()
> for (i in 1:1000)
> {
> a[i] <- sum(parSapply(cl, 1:15, get("+"), 2))
> }
> proc.time()-t1
> t2 <- proc.time()
See ?system.time
> a <- c()
> for (i in 1:1000)
> {
> a[i] <- sum(sapply(1:15, "+", 2))
> }
> proc.time()-t2


-- 
Brian D. Ripley,                  ripley at stats.ox.ac.uk
Professor of Applied Statistics,  http://www.stats.ox.ac.uk/~ripley/
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272866 (PA)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595