R help - Sep 2009 - more strange behavior of Revolution R 1.3.0

It runs more than twice as slowly using 8 core than using a single core
in inverting large matrix. Tested on 8 core Windows XP 64 machine. 

 

 
> n = 1000
> n.simu = 100 
> func1 = function()
+ {

+   x = rnorm(n*n)

+   dim(x)=c(n,n) 

+  y = solve(x)

+ }
> 
> setMKLthreads(1)
> system.time(for(i in 1:n.simu) func1())
   user  system elapsed 

  69.48    2.42   71.91 
> 
> setMKLthreads(8)
> system.time(for(i in 1:n.simu) func1())
   user  system elapsed 

 179.06   17.90   41.70

 

Jason Liao


	[[alternative HTML version deleted]]

Please discuss with REvolution support. Most of us do not have a version 
of REvolution R.

Uwe Ligges



Jason Liao wrote:
> It runs more than twice as slowly using 8 core than using a single core
> in inverting large matrix. Tested on 8 core Windows XP 64 machine. 
> 
>  
> 
>  
> 
>> n = 1000
> 
>> n.simu = 100 
> 
>> func1 = function()
> 
> + {
> 
> +   x = rnorm(n*n)
> 
> +   dim(x)=c(n,n) 
> 
> +  y = solve(x)
> 
> + }
> 
> 
>> setMKLthreads(1)
> 
>> system.time(for(i in 1:n.simu) func1())
> 
>    user  system elapsed 
> 
>   69.48    2.42   71.91 
> 
> 
>> setMKLthreads(8)
> 
>> system.time(for(i in 1:n.simu) func1())
> 
>    user  system elapsed 
> 
>  179.06   17.90   41.70
> 
>  
> 
> Jason Liao
> 
> 
> 	[[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.

The best place for questions specific to REvolution R is the REvolution
forums:
http://forums.revolution-computing.com/

(Note: due to anti-spam moderation, it may take a little while for your post
to appear.)

In this particular case, it looks to me like you're getting a significant
speedup after setting it to use 8 cores rather than 1 (elapsed time of 42 vs
72 seconds -- with threaded applications it's the wall clock that counts).
The fact that there is a big increase in the user (CPU) time indicates that
there is a lot of overhead for this problem in using multiple threads, but
nonetheless it's a net real-world benefit.

# David Smith

On Thu, Sep 24, 2009 at 7:13 AM, Jason Liao <JLIAO@hes.hmc.psu.edu> wrote:
> It runs more than twice as slowly using 8 core than using a single core
> in inverting large matrix. Tested on 8 core Windows XP 64 machine.
>
>
>
>
>
> > n = 1000
>
> > n.simu = 100
>
> > func1 = function()
>
> + {
>
> +   x = rnorm(n*n)
>
> +   dim(x)=c(n,n)
>
> +  y = solve(x)
>
> + }
>
> >
>
> > setMKLthreads(1)
>
> > system.time(for(i in 1:n.simu) func1())
>
>   user  system elapsed
>
>  69.48    2.42   71.91
>
> >
>
> > setMKLthreads(8)
>
> > system.time(for(i in 1:n.simu) func1())
>
>   user  system elapsed
>
>  179.06   17.90   41.70
>
>
>
> Jason Liao
>
>
>        [[alternative HTML version deleted]]
>
> ______________________________________________
> R-help@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.
>


-- 
David M Smith <david@revolution-computing.com>
Director of Community, REvolution Computing www.revolution-computing.com
Tel: +1 (206) 577-4778 x3203 (San Francisco, USA)

Check out our upcoming events schedule at
www.revolution-computing.com/events

	[[alternative HTML version deleted]]