07 May 2013


reproducible research, system administration, and git

Next post Previous post

Antlab workstation guidelines

System

Dell T7600 workstation running Ubuntu 12.04

  • Processor: Two four-core XEON E5-2609, 2.4GHz
  • Memory: 64GB DDR3 RDIMM 1600 ECC, 8x8GB
  • Hard drive: 4x3TB 3.5inch 7200 SATA 6Gb/s in RAID 6 for redundancy

Note that regardless of Dell stating that the system could run linux, no linux drivers were available for the installed PERC H310 SATA/SAS RAID controller. Pine Computers replaced the PERC card with an Adaptec AAC-RAID and were then able to install Ubuntu.

Running a job

If you will be running a long (more than 1 hour) or memory intensive (more than ~20 GB) job…

FIRST: check system status

The antlab system has 8 cores, so only 7 CPU-hungry long-term jobs can be run simultaneously before bad things happen. Before scheduling a job, check processor usage by typing in the command line:

top -n 3

and q to exit.

You will see a screen like this:

<img src="http://www.johnstantongeddes.org/assets/img/top-output.png" alt="plot" width="545" height="348">

which shows a list of the running processes (update every three seconds as specified by -n 3, sorted by CPU usage. This screenshot shows 6 jobs using over 100% CPU usage each! Percent usage is for 1 core, so a job using more than 100% is using more than one core.

If you see this…do not start a new job.

Only 6 jobs using >85% of CPU can be run simultaneously before bad things happen. Wait until there are free CPUs.

You can also check memory usage by hitting Shift-f and selecting n then press enter. The active processes will be ordered by memory usage and you can see what percent is still available. If total memory usage is over 80%…wait to start your job.

Then, queue your job:

To run a script, use the command

nohup nice -n 19 Rscript script.r &

where the nohup command specifies that your job will be run without hangups (if your terminal becomes disconnected) and nice specifies low priority to avoid swamping system resources, and the & means that the job runs in the background so you can continue working in the terminal. Replace Rscript with bash for a shell script, or python fore a python script, etc…

The output of the script will be concatenated to the file nohup.out in the working directory.

Note that while the nice command will help maintain system resources.

Stopping jobs

To kill a job that is running, use

kill -9 PID

where PID is the process ID listed from the top command. If you have a job running in the background (e.g. with & at the end of the line) you can use:

kill $!

which kills the last process executed in the background.

File management

File management for organization and reproducibility of results based on guidelines suggested on biostars and in Noble WS (2009) but modified for analyses performed primarily using R. Each top-level directory should contain a README file explaining the contents.

Organizational outline:

  • /home/projects directory, in which each subdirectory corresponds to a separate project
  • /home/data directory, which contains data used by more than one project
  • /home/lab directory, used to hold files miscellaneous lab files such as protocols, lab meeting papers and notes, etc

Within each project directory, I recommend the following directories

  • planning/ directory containing all the files related to the early phase of the project and miscellaneous file
  • data/ directory containing the different data I need to use, can include symbolic links to or from /home/data
  • src/ directory containing bash, python, other scripts
  • R/ directory containing R scripts, ideally with Roxygen documentation
  • results/ directory containing results
    • organize subdirectories with results by date such as “2013-05-18”, with each subdirectory containing a driver script with specific parameters for results. For R, this should be in R markdown format
  • doc/ directory with a subdirectory for each manuscript, etc
    • manuscript/
    • figures/
    • tables/

Symbolic linking

For large files, use symbolic linking as much as possible to save disk space. The command

ln -s </path/to/file/yourfile> <linkname>

will create a link to ‘yourfile’ in the current directory

File permissions

The chmod command sets file permissions

For a file that you want to protect from accidental changes or overwrites, use

chmod a-w <file>

where the a means all users and the -w means remove write permissions. Similarly

chmod ug+x <file> 

would mean that u user and g group members can execute a file (for example, run a script) or

chmod g-r <file>

would mean that group members cannot read a file, but you still can.

Removing files

The standard command to remove files is rm file but this is IRREVERSIBLE so should be used with caution. The trash-cli program is installed for ‘recycle bin’ funcionality, but I recommend version control as well (see next).

trash-put <file or directory>

and

restore-trash <file or directory>

to restore a file from your trash.

Version control

git should be used extensively, especially on files in src and R directories. More info coming…

Software installation

Requires administrator priviledges.

Software is either installed using the Ubuntu Software Center or apt-get.

Non-standard or newer versions of programs than are available through the software center are installed into /opt/software and symbolically linked from /usr/local/bin using

ln -s /opt/software/<target> <link>

so that they are on the $PATH and system-wide accessible. See for more information. May need to set permissions for group executability using

chmod ug+x <file>

Software can be installed in home directory for personal use at your own risk.

System maintenance

System and software updates will be performed the first Monday of every month, so if you are going to schedule long-running jobs around this time let me know if advance.

References

Noble WS (2009) A Quick Guide to Organizing Computational Biology Projects. PLoS Comput Biol 5(7): e1000424


Creative Commons Licence
This work is licensed under a Creative Commons Attribution 4.0 International License.