Using TACC

Many of the analyses performed in the lab require significant computational resources. For this reason, you should become familiar with using TACC resources to run your analyses. This will speed up computational time for time-intensive processes, like permutation tests or machine learning algorithms. All command line text is highlighted in bold. 

1. Request a TACC account via the TACC user portal
   1. https://portal.tacc.utexas.edu/
2. Ask Dr. Hussain to add you to an existing project allocation
   1. Typically, our project allocations allow us to run analyses on Stampede2.
   2. Review Stampede2's documentation to become familiar with its capabilities and hardware structure.
      1. https://portal.tacc.utexas.edu/user-guides/stampede2
   3. If a project allocation does not exist for your project, you can request one here: https://portal.tacc.utexas.edu/allocations-overview
3. TACC requires multi-factor authentication to log in. Choose your preferred method and set it up.
   1. Follow the directions at this link: https://portal.tacc.utexas.edu/tutorials/multifactor-authentication
4. If you are on a windows machine, download and install Putty. Putty allows you to remotely log into the Stampede2 supercomputer and run analyses on it.
   1. https://www.putty.org/
5. Open Putty (if on a windows machine) or terminal (if on a mac)
   1. Log into stampede2 using the following commands
      1. ssh -X username@stampede2.tacc.utexas.edu 
   2. Complete multi-factor authentication 
      1. enter your account password
      2. enter TACC passcode 
6. Copy data folder from Box to your TACC scratch directory
   1. rsync -a /local_directory/local_folder remote_username@stampede2.tacc.utexas.edu:/remote_directory/remote_folder
   2. Having trouble? Check out these resources: 
      1. https://phoenixnap.com/kb/linux-scp-command
      2. https://www.tecmint.com/rsync-local-remote-file-synchronization-commands/
7. Request compute nodes
   1. idev -N #nodes -t hh:mm:ss -p skx-normal
      1. here, #nodes = the number of compute nodes you want to request, and hh:mm:ss = the time you need those nodes to be available for
   2. If you are running parallel processes on stampede2, try to use SKX nodes instead of KNL nodes. These typically perform better. 
   3. Wait for resources to become available. This can take awhile, especially if you are requesting a large number of nodes.
8. Load matlab
   1. module load matlab 
   2. matlab 
      1. If matlab does not open or gives you a license error, contact the TACC helpdesk to ask them to activate your license on stampede2. Provide as much detail as you can when describing the problem. 
         1. https://portal.tacc.utexas.edu/tacc-consulting/-/consult/tickets/create
9. Run your analysis in matlab as you normally would, saving the data to your scratch folder
10. When finished, copy the newly saved data from your scratch folder back to Box 
    1. rsync -a remote_username@remote_host:/remote_directory/remote_folder/local_directory/local_folder
11. PRO TIP: When running a long analysis, caffeinate your machine to make sure it does not go to sleep and cause the TACC session to close
    1. PC: download and install this app
       1.  https://lifehacker.com/prevent-your-windows-pc-from-sleeping-with-caffeinated-1832271248
    2. Mac: open a new window in terminal
       1. type caffeinate