R/sys_perform_parallell.R

In jhollist/miscPackage: Package with JWHs Miscellaneous Functions

#' sys_perform
#' 
#' This function returns the time it takes to run the following: a big matrix
#' multiplication, a loop without pre-allocated mem, a loop with 
#' pre-allocated memory, and writing a csv to file. As I am not a computer 
#' scientist, I have no idea if these test are meaningful.  Just things I do a 
#' lot and wanted to see if they work faster on different machines. 
#' 
#' @param n numeric value indicating the number of times to run the test.  Default is 10
#' @param file Character string for file (and path) of output file to write results to
#' 
#' @examples
#' \dontrun{
#' #Following example will write to a shared file on ORD_public
#' #Linux
#' #You will need sudo rights and mount ORD_Public, like:
#' # $sudo mkdir /mnt/ord_public
#' # $sudo mount //V2626UMCEC501.aa.ad.epa.gov/ORD_Public$ -o username=aa\\YOURUSERNAME,
#' #            sec=ntlmsspi,uid=YOURUSERNAME /mnt/ord_public
#' 
#' #Linux Example then is
#' sys_perform(file='/mnt/ord_public/jhollist/R/epa_r_sys_performance.csv')
#' 
#' #Windows
#' 
#' #No need to map, the following should work
#' sys_perform(file='//aa.ad.epa.gov/ORD/ORD/DATA/Public/jhollist/R/epa_r_sys_performance.csv')
#' }
#' @export
#' @import snowfall
#' @import parallel
sys_perform_parallel <- function(n = 10, file = NULL, num_core = 1) {
    t1 <- proc.time()
    # validate number of cores
    if (num_core > parallel::detectCores()) {
        return(stop(paste("Your system only has", parallel::detectCores(), "cores available")))
    }
    
    # Set up single core/sequential
    if (num_core == 1) {
        sfInit(parallel = FALSE)
    }
    
    # Set up multiple cores/parallel
    if (num_core > 1) {
        sfInit(parallel = TRUE, cpus = num_core)
        sfLibrary(miscPackage)
    }
    
    # sys_perform Function to pass to sfLapply
    sf_sys_perform <- function(x) {
        out_sys_perform <- sys_perform(1)
        return(out_sys_perform)
    }
    
    # Run it
    result <- sfLapply(1:n, sf_sys_perform)
    
    # Summarize Results
    result <- data.frame(do.call("rbind", result))
    time <- proc.time() - t1
    result <- with(result, list(User.Name = unique(unlist(User.Name)), R.Version = unique(unlist(R.version)), 
        Machine.Name = unique(unlist(Machine.Name)), OS = unique(unlist(OS)), Memory.GB = unique(unlist(Memory.GB)), 
        Drive = unique(unlist(Drive)), Number.Runs = n, Number.Cores = num_core, matrix.multiply = (sum(unlist(matrix.multiply))/num_core)/n, 
        unallocated.loop = (sum(unlist(unallocated.loop))/num_core)/n, allocated.loop = (sum(unlist(allocated.loop))/num_core)/n, 
        write.csv = (sum(unlist(write.csv))/num_core)/n, avg.time = (sum(unlist(avg.time))/num_core)/n, 
        total.time = time[[3]], date = unlist(date)[1]))
    # Kill cluster
    sfStop()
    
    return(result)
    
}