Comparing Serial and Parallel Workflows

1
Comparing Serial and Parallel Workflows
Serial Analysis
Parallel Analysis

• Create Model
• Analyse Results

• Specify Host
• Authentication
• Create Model
• Poll for Status of Analysis
• Analyse Results

2
Serial R Script
library(sabreR) read the data hsblt-read.table(f
ile"/home/grosed/WorkBench/SabreRCourse/data/hsb.
tab") attach(hsb) create the
model sabre.model.1lt-sabre(mathach1,caseschool,
first.mass64,first.family"gaussian")
show the results sabre.model.1
3
Parallel (Grid) R Script
library(GROWL) mysession grid.session("User20"
,"caman8","lancs1.nw-grid.ac.uk", "lancs1.nw-grid.
ac.uk/jobmanager-sge","","") read the
data hsblt-read.table(file"/home/Shared/data/hsb.t
ab") attach(hsb) create the models sabre.model.1
lt-sabre.grid(mathach1,caseschool,
first.mass64,first.family"gaussian",
script.file"hsb1.sab",
log.file"hsb1.log",
data.file"hsb1.dat",
sabre.binary"/panfs/la/staging/users/dlab/SABRE5/
sabre_p.out",
sessionmysession,
processors1) print.sabre.grid(sabre.model.1)
4
Conceptual Differences

• Parallel Jobs on Remote Systems are Asynchronous
• Need to keep track of running jobs
• Need to leave R session and recover job
  information.
• Authentication
• Scheduling
• The same model may take different times to run
• Scheduling Policy and system architecture might
  effect choice for number of processors.