Page History

This is all about how to compute on coyote.

*{_}{+}Gaining access:+{_}*

I'm pretty sure that \[greg\] still helps out with this, you would probably need to ask him for access. Also, if you are working off campus, you need to log into your on-campus account first then ssh onto coyote. If you have questions about this, just ask me.


*{_}{+}What is available:+{_}*

+modules+

module add python/2.7.3
module add atlas/3.8.3
module add suitesparse/20110224
module add numpy/1.5.1
module add scipy/0.11.0
module add biopython

module add matplotlib/1.1.0

module add matlab

Not QIIME as far as I can tell.

*{_}{+}Interactive computing{+}{_}*:

I'm not sure how to work interactively. I've was able to get onto a node with:

qsub \-I \-l nodes=1

But when I tried to use matlab with (module add matlab) it didn't work (although it did work upon ssh'ing)


*{_}{+}Submitting multiple jobs:+{_}*


Before running the program below, make sure to load the following modules (I tried without and I got an error loading argparse):

module add python/2.7.3
module add atlas/3.8.3
module add suitesparse/20110224
module add biopython
module add java/1.6.0_21

Also, there are different timed queues, so make sure if you get this working that it submits to the right queue. From what I can tell there is short


From Mark and Chris (although I have yet to be able to make it work\- I can get the script to write but it doesn't run)\-


I've been using a script chris wrote which works pretty well:{color:blue}/home/csmillie/bin/{color}{color:blue}ssub{color}
*What it does*
It streamlines job submission. If you give it a list of commands,  it will (1) create scripts for them, and (2) submit them as a job array.  You can give it the list of commands as command line arguments or  through a pipe.
*Quick examples*
1. Submit a single command to the cluster {color:blue}ssub{color}{color:blue} "python /path/to/script.py > /path/to/output.txt"{color}
2. Submit multiple commands to the cluster (use semicolon separator) {color:blue}ssub{color}{color:blue} "python /path/to/script1.py; python /path/to/script2.py"{color}
3. Submit a list of commands to the cluster (newline separator) {color:blue}cat /list/of/commands.txt \| {color}{color:blue}ssub{color}
*Detailed example{*}{color:blue}* /home/csmillie/alm/mammammals/aln/95{color}/
In this directory, I have 12,352 fasta files I want to align. I can do this on 100 nodes quite easily:
1. First, I create a list of commands: {color:blue}for x in `ls \*fst`; do y=${color}{x%.*}{color:blue}; echo muscle \-in $x \-out $y.aln; done > commands.txt{color}
The output looks like this:
...
muscle \-in O95_9990.fst \-out O95_9990.aln
muscle \-in O95_9991.fst \-out O95_9991.aln
muscle \-in O95_9992.fst \-out O95_9992.aln
muscle \-in O95_9993.fst \-out O95_9993.aln
...
2. Then I submit these commands as a job array: {color:blue}cat commands.txt \| {color}{color:blue}ssub{color}
*How to configure it*
Copy it to your \~/bin (or wherever). Then edit the top of the script:{color:blue}uname{color} = your username {color:blue}tmpdir{color} = directory where scripts are created {color:blue}max_size{color} = number of nodes you want to use
*Other things{*}{color:black}It a{color}utomatically creates random filenames for your scripts and job arrays. These files are created in the directory specified by "{color:blue}tmpdir{color}" It can also submit individual scripts instead of a job array.

*{_}{+}How to run matlab:+{_}*

This is running interactively upon login. I'm not sure if there is a better way to do this.


ssh \-X username@coyote.mit.edu

module add matlab

matlab

\-A window should pop up and so you can use the matlab GUI.