Oscar: Sage

Loading and Launching Sage

1. OpenOnce theauthenticated Terminalto andOscar, use the following commands at the command line.

2.  Start an interactive job by using the interact command. This command can take additional parameters to extend the resources and time allotted to the node as well as the partition that the node operates on.

3. module load sage to load the most recent version of Sage.

4. If your work requires an older version of Sage, use module load sage/9.05  to load Sage. This command will return confirmation “module: loading 'sage/9.0'”.

3. Sage on Oscar usually has dependent modules. These will be output to your terminal when you load the Sage module similar to the example below.

[username@node ~]$ module load sage/9.0
module: loading 'sage/9.0'
module: sage: To use: module load gcc/8.3

4. Load any other dependent modules for Sage that were output in the above steps. These modules must be loaded in the specific order specified by the system when the Sage module is loaded. Loading them in a different order may cause Sage to throw errors.

For Sageversion 9.0: module load gcc/8.3

For Sage 8.7: module load gcc/8.3 sqlite/3.25.2 perl/5.24.1 libgd/2.2.5 mpfr/3.1.5 ffmpeg/4.0.1 imagemagick/7.0.7 texlive/2018

5.  sage to launch the Sage console within your Terminal window.

Sage on VNC

The Sage install may not always work properly on Oscar's VNC nodes due to the VNC nodes running older chipsets. The easiest way to remedy this issue is to run sage in an interactive job via the terminal in your VNC session.

Use the interact command with parameters for your specific job to start the interactive session, then load your modules and run the sage binary (steps 2-4 above).

interact -n 2 -m 32g -t 04:00:00 -f 'haswell|broadwell|skylake'

Installing Sage Packages Locally with --user

It is possible to install most Sage packages locally to your home folder by passing the the –user  parameter at the end of your install command. See below for example steps to install the Sage packages “surface_dynamics” and “sage-flatsurf”. In this example, we are loading both packages from git repositories, so we need to load the git module.

1.  module load git

2.  sage -pip install git+https://gitlab.com/videlec/surface_dynamics  –user

3.  sage -pip install git+https://github.com/videlec/sage-flatsurf –user

Using Sage with Batch Scripts

Thanks to Trevor Hyde from Summer@ICERM 2019 for these instructions.

One method for running computations with Sage on Oscar is to write a script and use the slurm batch scheduler to have Oscar run your script. This requires two pieces:

Example Batch Script

sage-batch.sh

#!/bin/bash
 
#SBATCH -J test_program
#SBATCH --array=0-9
#SBATCH -t 1:00:00
#SBATCH --mem=8G
 
#SBATCH -e data/<oscar-username>/test_output/test%a.err
#SBATCH -o data/<oscar-username>/test_output/test%a.out
 
module load sage/8.7
module load gcc/8.3 sqlite/3.25.2 perl/5.24.1 libgd/2.2.5 mpfr/3.1.5 ffmpeg/4.0.1 imagemagick/7.0.7 texlive/2018
 
sage test_program.sage $SLURM_ARRAY_TASK_ID

Everything after this in the script happens as if you typed it yourself onto the command line.

Example Sage Program

test_program.sage

import sys
 
def fun_math(message):
    print message
    sys.stdout.flush()
 
job_id = int(sys.argv[1])
fun_math('hi this is a test')
fun_math('my job id is' + str(job_id))

• In the Sage program, you first define all of your functions and then you include the code you want to run.

• Import Import sys  so you can access the array parameter passed to your function from the node. This is accessed in this case by by sys.argv[1]. Make sure you explicitly coerce to be an integer if you want to use it as an integer; it's a string by default.

• The output of the the print  command is appended to the the .out  file for this node as a new line.

• Notice the line line sys.stdout.flush()  included in the function. This makes the program immediately send whatever output it has to the output file when called. Otherwise the program won’t output output anything  until it has completely finished running. If each node is running 100 potentially long computations and it finishes the first 99 but then times out on the 100th computation, and you don’t include any any sys.stdout.flush()  commands, everything will be lost when time runs out.

Submitting the Batch Job