From MCS IT Wiki
The crucial bits you need to know when working in our linux environment.
- /sandbox is scratch space intended for fast, local data access. It is not backed up. Pretend it's a ramdisk.
- login machines (login.mcs.anl.gov) are interactive login machines. People rely on them for interactive session. Do not monopolize CPU on these machines. See this section for further information on which machines to use.
- Don't monopolize CPU on anyone's desktop other than your own without explicit permission. If we are asked to kill a process on someone's desktop by that person, we will do so.
2012 MCS Green Build
Commercial Software Available
In order to use our site licensed software you will need to add a line with the appropriate "softkey" to the file <yourusername>/.soft. After editing that file run the command "resoft" to apply the changes to your current shell.
|32 bit||64 bit||Name||softkey|
|X||X||Absoft Fortran 11||+absoft|
|X||X||NAG Fortran 5.3||+nagfor|
|X||X||Portland Group Compiler 12.10||+pgi|
Citrix Receiver Installation (Dayforce)
Argonne recently launched Dayforce, a new time management system for lab wide use. To use this application, you need to install the Citrix Receiver. This should now be available on the MCS green build by default.
If you find yourself needing to install this on a personal Linux machine (either 32 or 64 bit variants), you can find Argonne's instructions here. Please note that only version 12.1 of the receiver is currently supported by Argonne.
You will notice, however, that the package supplied from Citrix is broken in many ways and several extra steps are needed to install this properly. We in MCS created a package (for 64 bit Debian based systems) that will take care of all of this for you. To install onto a 64 bit system you can follow these steps:
1) Download our package to a temporary directory. The package is available here: Icaclient_12.1.0_amd64_v2.deb
2) Install the package dependencies:
sudo apt-get install -y libmotif4:i386 libxpm4:i386 nspluginwrapper flashplugin-nonfree-extrasound
3) Install the package:
sudo dpkg -i /path/to/file/Icaclient_12.1.0_amd64_v2.deb
If you get a complaint about needing more dependencies, run:
sudo apt-get -f install
then run the client installer again.
4) You should now be able to launch the application. Browse to http://appgateway.anl.gov and you'll be prompted to login.
If you see a notice to Install the Citrix application, you can click on this and then close the download page it sends you too. We've already installed this. Navigate back to the previous page and you should now be able to launch the Dayforce application. It may take some time.
If you have questions, please e-mail firstname.lastname@example.org
Using CUDA in MCS
CUDA is installed for general use within MCS. Any machine with a CUDA enabled GPU can run this. You can find a list of supported GPUs here. You'll also need to make sure you have the appropriate NVIDIA driver installed. If this is an MCS machine, you can contact email@example.com to make sure this is the case.
Once you've determined you can run CUDA, you need to enable it in your environment by adding a "sofkey". If you have an MCS home directory, you can edit the .soft file in your home folder (<your username>/.soft) with your favorite text editor. To get the current version of CUDA, on a new line at the end of the file add:
The current version we are supporting is 5.0 and you'll get this with the softkey listed above. As of 12/13/13 we also have version 5.5 available and this can be added by instead using the softkey:
To apply the changes to your current shell after editing that file run the command:
Once it has been determined that you have a CUDA enabled GPU, the correct driver and have added the softkey to your environment, you should now be able to use CUDA.
You can check with version of the compiler is in use with the command:
Pre-built samples are also packaged with CUDA to test the functionality. Running a sample and getting the correct output will verify that you meet all of the requirements to actually use CUDA. If you run the below command with version 5.5 and get real output back, you are good to go:
PGI CUDA Fortran
PGI and NVIDIA defined CUDA Fortran to enable CUDA programming directly in Fortran. CUDA Fortran is a small set of extensions to Fortran that supports and is built upon the CUDA computing architecture.
You can access these features by enabling both the CUDA softkey and the PGI softkey in your .soft file. You should already have CUDA enabled as described above, so simply add the PGI key.
and the once again reload your shell with this command:
The Portland Group has provided a sample file to test that this is working correctly. You can download their sample here.
Once you have this file, run:
pgfortran -fast matmul.CUF
and then if run the output file. If it builds and runs with no errors, you are all set.
The .CUF extension tells PGI to run it as CUDA compiled code or you can have a standard .f extension and add this flag when compiling instead: -Mcuda