Graduate Computing Workshop exercises.

Part 1: Uploading Workshop Files to ACI

Method 1. Copy using Git

1. Go to your work directory:

cd ~/work

2. Clone repository using:

git clone git@github.com:benlee55psu/workshop.git

Method 2. Copy Directly to ACI

1. Go to the github site https://github.com/benlee55psu/workshop
2. Click on the green "clone or download" button. Download ZIP.
3. Unzip files and note the local directory of the workshop directory
4. Copy workshop directory to ACI. Open the command line on your local machine. Then type in the following:

cd ~/<LocalDirectory>

scp -r ../workshop <userid>@datamgr.aci.ics.psu.edu:~/work/

Part 2: Installing packages + Rmpi

1. Go to the workshop directory on ACI

cd ~/work/workshop/

2. Install necessary packages. Type the following into the command line:

module load gcc/5.3.1

module load openmpi/1.10.1

Rscript installPackages.R

(Ignore Open Fabrics and MXM errors. This happens when you are running tests on the login node, which cannot access the MPI enviroment.)

Part 3: Exercises

Exercise 1 - Run a Toy Example on the Command Line

This takes ~1.7 minutes

1. Go to the proper directory

cd ~/work/workshop/exercise1/

2. Run job using Rscript

Rscript exercise1.R

3. Copy Files to your machine scp <userid>@datamgr.aci.ics.psu.edu:~/workshop/exercise1/. <LocalDirectory>/workshop/exercise1/.

4. Summary

Exercise 2 - Run Toy Example using the Batch Scheduling System

This takes ~2 minutes

1. Go to the proper directory

cd ~/work/workshop/exercise2/

2. submit the PBS script

qsub exercise2.PBS

qstat -a -u <userid>

3. Copy Files to your machine

scp <userid>@datamgr.aci.ics.psu.edu:~/workshop/exercise2/. <LocalDirectory>/workshop/exercise2/.

Exercise 3 - Run Toy Example using 10 processors Simultaneously

This takes ~15 seconds vs. ~2 minutes for Exercise 2

1. Go to the proper directory

cd ~/work/workshop/exercise3/

2. submit the PBS script

qsub exercise3.PBS

qstat -a -u <userid>

3. Copy Files to your machine

scp <userid>@datamgr.aci.ics.psu.edu:~/workshop/exercise3/. <LocalDirectory>/workshop/exercise3/.

Exercise 4 - Run Toy Example using 40 processors Simultaneously

Notes

• Please make sure Rmpi is installed (Installation instructions) before running exercise 4.
• FOR JOBS USING <100 PROCESSORS (<5 NODES), OPEN QUEUE IS AN OPTION.
• FOR JOBS USING 100+ PROCESSORS (5+ NODES), USE AN ALLOCATION.
• This takes ~31 seconds vs.~15 seconds for Exercise 3 vs. ~2 minutes for Exercise 2
• Slower times due to communication costs.

1. Go to the proper directory

cd ~/work/workshop/exercise4/

2. Submit the PBS script

qsub exercise4.PBS

qstat -a -u <userid>

3. Copy Files to your machine

scp <userid>@datamgr.aci.ics.psu.edu:~/workshop/exercise4/. <LocalDirectory>/workshop/exercise4/.

Exercise 5 - Run Toy Example by iteratively submitting 20 jobs with 1 processor per job

1. Go to the proper directory

cd ~/work/workshop/exercise5/

2. run the following bash command to send 20 separate PBS jobs

bash exercise5.bashrc

qstat -a -u <userid>

3. Copy Files to your machine

scp <userid>@datamgr.aci.ics.psu.edu:~/workshop/exercise5/. <LocalDirectory>/workshop/exercise5/.