-
Notifications
You must be signed in to change notification settings - Fork 1
Using the cod nodes
This document describes how to use the high-performance compute nodes of the cod group at CEES.
See here.
If you intend to use one of the nodes for an extended period of time, please send an email to the cod-nodes mailing list.
To log in type the following command into your command line interface (e.g. Terminal for Mac users)
ssh [email protected]
When you are on the UiO network it is enough to write ssh cod5
Check that you are a member of the group 'seq454' by simply typing:
groups
If 'seq454' s not listed, please contact Lex Nederbragt.
First thing you should do before working on the cod nodes is to create a bash script in your home directory (/usit/abel/u1/username), which is the default area after logging in. This is done by using a terminal-based text editor (e.g. nano, emacs etc.)
nano .bash_login
This command opens the text editor where you copy paste the following text:
export PATH=/projects/cees/bin:/projects/cees/scripts:$PATH umask 0002 module use --append /projects/cees/bin/modules
This is necessary to:
- have access to the commonly used programs and scripts (export PATH=/projects/cees/bin:/projects/cees/scripts:$PATH)
- help with automatically setting permissions to new files and folders (umask 0002)
- use our own modules (module use /projects/cees/bin/modules) - see [https://wiki.uio.no/mn/bio/cees-bioinf/index.php/User_manual_cod_nodes#Software_not_installed_yet below]. The '--append' ensures that is the same module is available already on abel, the abel one is chosen, not the local one.
Save the file and exit the terminal-based text editor.
'''Advanced tip''': To get a fancy prompt add this to your .bash_login file:
magenta=$( tput setaf 5) cyan=$( tput setaf 6) green=$( tput setaf 2) yellow=$( tput setaf 3) reset=$( tput sgr0) bold=$( tput bold) export PS1="\[$magenta\]\t\[\$reset\]-\[\$cyan\]\u\[\$reset\]@\[\$green\]\h:\[\$reset\]\[\$yellow\]\w\[\$reset\]\$ "
Advice on how to set up a folder for your project in /projects/cees/in_progress is given here
On the cod nodes
type
df -h
Look for the /node/work and /node/work partitions
For the /projects/cees area
type
ssh abel.uio.no dusage -p cees
If you have recurrent problems of ‘broken pipe’ when working on the cod-nodes or abel (meaning you need to log in again after just a couple minutes of inactivity), here is a solution.
Locallly add the following to ~/.ssh/config:
HOSTS* ServerAliveInterval 60
With this setup the connection will be poked every 60 sec, and will therefore be kept alive. Thanks to Tore Oldeide Elgvin for the tip.
All nodes have 'abel' (the UiO supercomputer cluster) disks, and your abel home area mounted to them. So, all the files located in /projects are available on the cod nodes, see below. In addition, the nodes have local discs, currently:
/node/data --> for permanent files, e.g input to your program
/node/work --> working area for your programs - see below for details.
-
do not use your abel home area, you only have 200 GB and you are not meant to share data there with others, e.g. your colleagues
-
data on /projects/cees is backed up by USIT, but NOT data on /node/data and /node/work
-
reading and writing data to and from /node/data and /node/work will be much faster and efficient than to /projects/cees
-
DO NOT USE /projects/cees for data needed as (medium to large) analyses; use /node/work on the cod nodes, $SCRATCH for SLURM jobs, or - if we decide this - /work on Abel
This leads to the following '''strategy for how to choose which disk to use''': -
for something short and quick, eg. less, tar, you can directly work on data in /projects/cees
-
for a long running program, or one that generates a lot of data over a long time, use the locally attached /data and /work
-
once the long running job is done, you can move the data you want to keep to /projects/cees
-
NOTE having your program write a lot over a long time to a file on /projects/cees causes problems for the backup system, as the file may be changed during backup
-
NOTE use compression (gzip, pbzip2) where possible!
-
for long-term storage of data you do not need regular access to, please use the norstore allocation, see below
- long term storage of data you may need to access: Nird disk (time to recover the data: intermediate, rsync to working area )
- data from finished publications: appropriate database (e.g. genbank, SRA), datadryad, figshare, or Nird archive
- storage of project data that you need to access occasionally/regularly, finished analyses: /projects/cees/in_progress
See this page.
Command to get into Nird:
ssh login.nird.sigma2.no
- NOTE that you need a separate password for nird, you cannot use your UiO password. If you don't know your Nird passwrod, you can reset it at https://www.metacenter.no/
- when you log in, you will be in the folder /nird/home/username
- our data is stored in /projects/NS9003K
- folder there are:
454data, runsIllumina, runsPacbio --> from the Norwegian Sequencing Centre, do not touch
projects --> where you can store your files
In /projects/NS9003K/projects, please use the same foldername as you use on Abel in /projects/cees/in_progress. Add clear README's
NOTE in general it is wise to copy data first, then delete the original. This prevent accidental data loss. Do not use the 'mv' command for big files!
Use ''rsync'', it preserves permissions and timestamps, and allows for finishing an interrupted copy job without having to copy every file again.
'''TIP''' use 'screen' (Only possible with option 2)
'''TIP''' see note on tarballs and md5 sums below
''Option 1'': when you are logged in on Nird
cd /projects/NS9003K/projects/path/to/yourfolder rsync -av cod5.uio.no:/projects/cees/in_progress/path/to/folder_to_copy .
NOTE the '.' at the end NOTE adding a trailing slash '/' to the folder_to_copy will only copy its content, not the whole folder!
''Option 2'': when you are logged in on abel/cod nodes
cd /projects/cees/in_progress/path/where/folder_to_copy/is rsync -av folder_to_copy login.nird.sigma2.no:/projects/NS9003K/projects/path/to/yourfolder
NOTE adding a trailing slash '/' to the folder_to_copy will only copy its content, not the whole folder!
tar -cvzf filename.tgz your_folder
This will collect and compress your_folder, with all files and folders in it (i.e., recursively) into one big file. NOTE please add a clear README to your tarball! NOTE this may take a long time, use 'screen'!
'''TIP''' generate an md5 checksum:
md5sum filename.tgz > filename.tgz.md5
This allows for checking whether two files are identical (really and completely). The program generates a long, unique string that is different for each file (one byte difference creates an entirely different string).
NOTE this takes a long time for large files, use 'screen'.
Once your file is on Nird, run the same md5sum command and compare the output. Alternatively, run this command:
md5sum --check filename.tgz.md5
A large number of programs is already installed on abel, and can also be used when logged in to the cod nodes. These programs have been installed and are maintained by people at USIT and are available for all users of abel (not just CEES). This software is available through the 'module' system, which is explained in more detail in (this manual)[https://www.uio.no/english/services/it/research/hpc/abel/help/user-guide/modules.html]. If you're logged in to abel or the cod nodes, type
module availand you'll see a list of all the available programs (it will take a few seconds until the list is loaded).
As USIT people are too busy to answer every request for software installation on abel, an increasing amount of software has been installed by CEES users. This software lives in /projects/cees/bin, and in order to keep things organized, we also follow the module system. If you have properly set up your environment above, this type of software also shows up in the list that you see when typing module avail
. You'll see it at the top of the list, under the heading '--- /projects/cees/bin/modules ---', while the modules maintained by USIT appear below, under heading '--- /cluster/etc/modulefiles ---'. Some software in /projects/cees/bin is not yet accessible through modules, but only cause we haven't gotten around to that yet. If you need to use software that's installed in /projects/cees/bin, but has no module associated with it, please contact Lex or Micha, and we'll prepare one.
If you need to use software that's not installed on abel and the cod nodes yet, there are two ways to make it available. If you're not the only one who needs this software and it is likely to be used by many researchers, it might be worth asking people at USIT whether they could install it for you. To do so, send an email to [mailto:[email protected] [email protected]] with a polite request, and they'll see what they can do. If you seem to be the only one interested in this software, or if USIT people are too busy to care, you can install the program yourself, in /projects/cees/bin/, and with the corresponding module. To do so, please see the instructions given here.
'''Note:''' all nodes run Red Hat Enterprise Linux version 6, except cod2, which uses version 5 (normally, you will not notice this)
If you are used to submit jobs through a slurm script, this will not work on the cod nodes. Here you'll have to give the command directly on the command line.
You can use a job script: collect a bunch of commands and put them in an executable file. Run the command with
source yourcommands.sh
If you are starting a long running job you cannot close the terminal window because the job will be cancelled. Instead create a 'screen' in the terminal window from which you start to run the job: type
screen
You now started a 'new' terminal. Start your job in this new window. After you have started your job you can detach from this screen by pressing ctrl-a-d (The CTRL key with the 'a' key, followed by the 'd' key). Now you're back in the terminal where you started. You can close this terminal or even the computer and the 'new' terminal will still exist and continue to run your job. You can start multiple hidden windows by just repeating this procedure. If you want to get back into the now hidden screen type
screen -rd
If you have multiple hidden windows this command will give you a list of the windows available. To choose one of the windows just add the window-number after the command.
screen -rd 45142.pts-15.cod4
When your job is finished you need to close the hidden window by typing exit instead of ctrl-a-d
'''Advanced tip''': to make the environment 'inside' the screen more identical to your ususal shell, make a file in your home directory called .screenrc (i.e. /usit/abel/u1/username/.screenrc), with the following content:
shell -$SHELL
Back to the CEES-HPC wiki home page