Index of /examples/r/examples/io_scratch

Icon  Name                    Last modified      Size  Description
[DIR] Parent Directory - [   ] Rjob 15-Mar-2014 22:34 34 [   ] _data.csv 04-Feb-2016 14:32 4.8M [   ] io_scratch.R 15-Mar-2014 22:01 2.2K

Using scratch space for reading and writing files

The directory contains R script io_scratch.R
The script demonstrates an effective way to use on node (local) scratch space for reading and writing large datasets.


# Enviroment variable TMPDIR is created when the job starts,
# It contains the path to the temporary directory created in /scratch space for this job
# To get the value of the enviroment variable use Sys.getenv() command:
tmpdir <- Sys.getenv("TMPDIR")

# User can create his own subdirectory using for example his user ID:
tmpdir <- paste("/scratch/",Sys.getenv("USER"),sep="")
system(paste("mkdir",tmpdir))

#Create some data to write to a file
t <- rnorm( 100000, 50, 10 )
p <- rnorm(100000,998,50)
prec <- rbinom(100000,1,0.2)
dt <- data.frame(temperature=t,pressure=p,precipitation=prec)

# Measure the time writing dataset to the project space (over the network) and local scratch directory
system.time(write.csv(dt, file="_data.csv"))
 user  system elapsed
0.481   0.006   0.729
system.time(write.csv(dt, file=paste(tmpdir,"/_data.csv",sep="")))
 user  system elapsed
0.440   0.009   0.450

# In some cases the data has to be written one record at a time
out.dt <- function(infile, line){
 write.table(line, file=infile,append=TRUE, col.names =FALSE, row.names=FALSE, sep=",")
}

# The most inefficient way: write one record at a time appending it to the end of the file.
# Writing data this way involves execution of 4 commands for every record to be written:
# 1. openning the file
# 2. moving to the end of the file
# 3. writing a line
# 4. closing the file
infile <- "_data.csv"
system.time(for(i in 1:10000) out.dt(infile,dt[i,]))
 user  system elapsed
6.194   2.179  15.146

# Writing to the local scratch directory will be significantly faster.
infile <- paste(tmpdir,"/_data.csv",sep="")
system.time(for(i in 1:10000) out.dt(infile,dt[i,]))
 user  system elapsed
6.007   1.677   7.711

# The best way to output the data to the file one record at a time is
# to open a file (preferably in the local scratch directory)
# and then use the connection to this file to write the records.
# At the end the connection to the file needs to be closed.
infile <- paste(tmpdir,"/_data.csv",sep="")
file.conn <- file(infile, open="w")
system.time(for(i in 1:10000) out.dt(file.conn,dt[i,]))
 user  system elapsed
5.278   1.472   6.771
close(file.conn)

#Clean-up
unlink("_data.csv")
unlink(paste(tmpdir,"/_data.csv",sep=""))

This script can be copied from
/project/scv/examples/r/examples/io_scratch/

Interactive execution:

The script can be executed using source() funcion inside R enviroment.
It can also be executed using R or Rscript commands:
scc1% R CMD BATCH io_scratch.R
or
scc1% Rscript io_scratch.R

Submit R job:

To submit this script to the batch system, execute:
scc1% qsub -b y Rscript io_scratch.R

This directory contains a simple Rjob script that when submitted to the batch system, will execute the io_scratch.R script. To submit the job, type the following command at the SCC prompt:
scc1% qsub Rjob
Users who are members of any Medical Campus project must also include their project group name:
scc1% qsub -P scv Rjob

Contact Information:

Katia Oleinik: koleinik@bu.edu

Note: Research Computing Services (RCS) example programs are provided "as is" without any warranty of any kind. The user assumes the entire risk of quality, performance, and repair of any defects. You are encouraged to copy and modify any of the given examples for your own use.