[R-sig-ME] glmmTMB, OpenMP and Ubuntu

[Ben Bolker]

  [keeping r-sig-mixed-models in Cc]

    Don't know exactly how TMB::openmp() works ... ?
    The way I would check if it's running is to see what happens to CPU 
load when you run your model ...

On 9/7/20 5:02 PM, mega-x using gmx.net wrote:
> Dear Ben,
> 
> thank you for your quick response. I am still not sure if OpenMP is actually
> running. The command TMB::openmp() detects all 8 cores on my Linux Server.
> However, it also detects all 4 cores of my personal Windows machine that
> definitely does not support OpenMP.


> 
> The wording in the vignette sounded like you actively have to do something
> with your R installation. Your comment suggest that OpenMP is already part
> of the OS? I am still very confused here.
> 
> I am detecting a very slight improvement by switching from 1 to 8 cores on
> my Linux system (using the code of the vignette). I will do some more
> testing today including running the same model on a 96 cores cluster.
> 
> Cheers,
> 
> Xaver
> 
> -----Original Message-----
> From: R-sig-mixed-models <r-sig-mixed-models-bounces using r-project.org> On
> Behalf Of Ben Bolker
> Sent: Montag, 7. September 2020 20:48
> To: r-sig-mixed-models using r-project.org
> Subject: Re: [R-sig-ME] glmmTMB, OpenMP and Ubuntu
> 
>      Try running TMB::openmp() . If you already have R installed with OpenMP
> support, that should tell you how many cores OpenMP is configured to run
> with.
>      In general R automatically gets built with OpenMP support if a given
> OS/compiler combination supports it; I don't know much about how to
> troubleshoot that if it *doesn't* happen automatically, but there seems to
> be a little information  here
> 
> https://cran.r-project.org/web/packages/ProFit/vignettes/ProFit-OpenCL-OpenM
> P.html
> 
>     After it's installed, you could try running some fairly big models with
> glmmTMB(..., control=glmmTMBControl(parallel=xx)) where xx is either 1 or >1
> (NULL, the default, automatically sets the max number of OpenMP threads to
> the number of available cores on my machine).  Use
> system.time() to check the elapsed time (if you want to be more systematic
> and your examples are fast enough to run several times, you can use one of
> the available R benchmarking package).  I would also suggest using 'top' or
> some system monitoring GUI to check the CPU load while your job is running
> ... the expectation is that the CPU % listed for your R process will reflect
> the number of OpenMP threads being used.
> 
>     If you encounter anything particularly interesting you could report it at
> https://github.com/glmmTMB/glmmTMB/issues/620 ...
> 
> 
> On 9/7/20 5:56 AM, mega-x using gmx.net wrote:
>> Dear all,
>>
>>
>>
>> I am currently fitting a complex model with glmmTMB that involves random
>> effects and covariance structures (spatiotemporal autocorrelation). I
> would
>> like to try the experimental parallel optimization feature
>> <https://cran.r-project.org/web/packages/glmmTMB/vignettes/parallel.html>
>> to reduce computation time. To do this I've setup a virtual machine
> running
>> Ubuntu 20.04 on 16+ cores. However, I found no information in how to setup
>> OpenMP, R and glmmTMB. Is there a step by step guide that I can follow?
>>
>>
>>
>> The Vignette states the following:
>>
>> "If your OS supports OpenMP parallelization and R was installed using
>> OpenMP, glmmTMB will automatically pick up the OpenMP flags from R's
>> Makevars and compile the C++ model with OpenMP support."
>>
>> I was not able to find any information on how to install R using OpenMP.
>>
>>
>>
>> Best,
>>
>>
>>
>> Xaver
>>
>>
>> 	[[alternative HTML version deleted]]
>>
>> _______________________________________________
>> R-sig-mixed-models using r-project.org mailing list
>> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>>
> 
> _______________________________________________
> R-sig-mixed-models using r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>