Sets the "iscream.threads" option to n_threads. To see how many threads
you have available see ?get_threads().
Details
iscream uses OpenMP to parallelize certain functions. You can use
as many threads as are available to you on your system to varying degrees of
performance improvements. The get_threads() function uses
parallelly::availableCores() to report the number of available threads.
Although OpenMP can detect the number of available cores, on high
preformance computers (HPCs) with resource allocating job schedulers like
SLURM, OpenMP may detect all available threads across the HPC and not limit
itself to the cores that were allocated to you by the scheduler. If your
system administrator has not set up any limits, this may result in your job
taking resources from other jobs. If there are limits, trying to use more
threads that those available will reduce iscream's performance. Job
schedulers will typically have an environment variable (e.g.
SLURM_CPUS_ON_NODE with SLURM) that gives you the actual number of
available cores. Further, on hyperthreaded systems, this count may be double
that of the available processors. Using hyperthreading does not guarantee
any performance improvement - it may be better to set the number of threads
to half the reported number. parallelly::availableCores() takes HPC
scheduler/CRAN/Bioconductor limits into account when reporting the number of
available threads but it may not reliably report hyperthreading ('system' or
'nproc'). To set the number of threads without having to call
set_threads() in every session, put
in your .Rprofile See help('Rprofile') for information on startup options.
Functions currently using OpenMP:
Examples
(ncores <- parallelly::availableCores())
#> system
#> 4
if (FALSE) { # \dontrun{
set_threads(ncores)
} # }