Skip to content

Warning: This page is waiting for an update, the Brazos cluster is no longer in operation.

Please contact Mitchcomp help if you have any questions.

This guide is meant to help users set up and safely utilize interactive sessions in the Brazos cluster. There are limitations on this method, which are outlined below. If you need help contact us: Jorge D. Morales.

Interactive Use

You may request SLURM interactive sessions on a compute node via two simple commands: sintr or srun. The first one is a simpler version but doesn’t allow all the possibilities, we recommend the second one. You can see documentation for it in the SLURM documentation for srun. Look for the requirements of your session (i.e. memory, wall-time, to make sure you get the resources you need).

You can start an interactive bash session:

srun -p interactive --pty --mem 1900 /bin/bash

This is what this command is doing:

  • srun : basic SLURM submission command
  • -p interactive : choosing the ‘interactive’ partition (queue)
  • –pty : allows the session to start like a standard terminal
  • –mem 1900 : requesting 1.9GB of memory
  • /bin/bash : starting a bash session

Or if you have a script or program to run, it may be easier to simply run it in interactive mode, for example to run Matlab, do the following (after initializing the required modules and environment):

srun -p interactive --pty --x11=first --mem 8000 matlab

In this case we are adding:

  • –x11=first : allowing x11 windows to operate between compute and login nodes
  • –mem 8000 : requesting more memory (Matlab is more demanding)
  • matlab : of course after setting up the environment/modules

More details about running Matlab in Brazos

Most likely you will be using cdmstools, make sure you set it up properly ahead, by following the CDMSTools Guide.

Remember to source root, by adding the following lines in your .bash_profile file and sourcing it.
>$ emacs ~/.bash_profile – Edit the file and add the following lines
## ROOT CONFIGURATION VARIABLES ##
source /home/hepxadmin/root/bin/thisroot.sh
export ROOTSYS=/home/hepxadmin/root
export PATH=$PATH:/home/hepxadmin/root/bin
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/home/hepxadmin/root/lib
module load gcc/4.8.2 openblas
module load matlab

At this point you can request the interactive session, and begin using matlab:

srun -p interactive --pty --x11=first --mem 8000 matlab

For additional instructions and troubleshooting see the official Brazos guide to using matlab.

WARNING! Matlab Bug (in R2014a)

If you use Matlab 2014a, and this issue has been observed in our cluster, there is a chance that you get an error while trying to read RQs (after CAP_init of course). This issue is related to a Matlab bug (supposedly fixed in Matlab R2014b), which makes some of the libraries (especially the libblas) be unaccessible. It is not clear why this happens in our cluster, but there is a trick to prevent it from happening. The library especially fails in CAP functions when trying to multiply matrices (yeah, kinda basic…) but if you do a matrix multiplication before initializing CAP then things will run smoothly! Do the following matlab commands, just after starting matlab:

  • $> m1 = rand(2000,1000);
  • $> m2 = rand(1000,1);
  • $> m1*m2
  • >$ CAP_init(”,0) – just initialize CAP normally after the matrix multiplication


While this gets fixed this ‘trick’ seems to work fine.

Troubleshooting

There are many reasons a connection to Matlab might fail. For starters, at the time of writing, Brazos only has access to 3 matlab licenses. Check your timeout error, if it mentions anything about licensing, it’s possible each license is unavailable, and you may need to try again later.

Try connecting to login03 on Brazos. Login03 is a dedicated node for moderate cpu/memory usage. For things like quick calculations, idle stuff, debugging, etc. Login01 and login02 are just for submitting and checking the status of your jobs. In login03 there’s around 64GB of memory.

Try reducing your memory request. If your request seems to be taking an unusually long time to process (check the status of your job using squeue), it’s possible there’s not enough memory available on the node you’ve requested. Speaking of which…

Try requesting a specific node. For example:

srun -p stakeholder-4g --pty --x11=first --mem 8000 matlab

For a complete list of Brazos nodes click here.

For the official SLURM documentation on srun click here.

For the official Brazos documentation on using srun click here.

If all else fails, send an email to Brazos support at BRAZOS-HELP@listserv.tamu.edu with the command you’re running and as much information as possible.