run_on_cluster: Framework for running simulations on a cluster computing...

View source: R/run_on_cluster.R

run_on_clusterR Documentation

Framework for running simulations on a cluster computing system

Description

This function serves a scaffold for running simulations in parallel on a cluster computing system. It acts as a wrapper for the code in your simulation script, organizing the code into three sections, labeled "first" (code that is run once at the start of the simulation, e.g. setting simulation levels), "main" (the simulation script, which is run repeatedly), and "last" (code to combine and summarize simulation results). This function interacts with cluster job scheduler software (e.g. Slurm or Oracle Grid Engine) to divide parallel tasks over cluster nodes. See https://avi-kenny.github.io/SimEngine/parallelization/ for an overview of how cluster parallelization works in SimEngine.

Usage

run_on_cluster(first, main, last, cluster_config)

Arguments

first

Code to run at the start of a simulation. This should be a block of code enclosed by curly braces that that creates a simulation object. Put everything you need in the simulation object, since global variables declared in this block will not be available when the 'main' and 'last' code blocks run.

main

Code that will run for every simulation replicate. This should be a block of code enclosed by curly braces that includes a call to run. This code block will have access to the simulation object you created in the 'first' code block, but any changes made here to the simulation object will not be saved.

last

Code that will run after all simulation replicates have been run. This should be a block of code enclosed by curly braces that takes your simulation object (which at this point will contain your results) and do something with it, such as display your results on a graph.

cluster_config

A list of configuration options. You must specify either js (the job scheduler you are using) or tid_var (the name of the environment variable that your task ID is stored in). Run js_support() to see a list of job schedulers that are currently supported. You can optionally also specify dir, which is a character string representing a path to a directory; this directory will serve as your working directory and hold your simulation object, temporary SimEngine objects, and simulation results (this defaults to the working directory of the R script that contains your simulation code).

Examples

## Not run: 
# The following is a toy simulation that could be run on a cluster computing
# environment. It runs 10 replicates of 2 simulation levels as 20 separate
# cluster jobs, and then summarizes the results. This function is designed to
# be used in conjunction with cluster job scheduler software (e.g. Slurm or
# Oracle Grid Engine). We include both the R code as well as sample BASH code
# for running the simulation using Oracle Grid Engine.

# This code is saved in a file called my_simulation.R
library(SimEngine)
run_on_cluster(

  first = {
    sim <- new_sim()
    create_data <- function(n) { rnorm(n) }
    sim %<>% set_script(function() {
      data <- create_data(L$n)
      return(mean(data))
    })
    sim %<>% set_levels(n=c(100,1000))
    sim %<>% set_config(num_sim=10)
  },

  main = {
    sim %<>% run()
  },

  last = {
    sim %<>% summarize()
  },

  cluster_config = list(js="ge")

)

# This code is saved in a file called run_sim.sh
# #!/bin/bash
# Rscript my_simulation.R

# The following lines of code are run on the cluster head node.
# qsub -v sim_run='first' run_sim.sh
# qsub -v sim_run='main' -t 1-20 -hold_jid 101 run_sim.sh
# qsub -v sim_run='last' -hold_jid 102 run_sim.sh

## End(Not run)

Avi-Kenny/SimEngine documentation built on June 23, 2022, 11:09 a.m.