Getting an account
To login to the cluster, you need an SSH client of some sort. If you are using a linux or unix based system, there is most likely one already available to you in a shell, and you can get to your account very quickly.
If you are having problems connecting to your account, please let the HPC Staff know by submitting a Bugzilla Request.
Linux / Unix
So, here is how you would go about logging in via a linux or unix account:
test@puppy:~$ ssh email@example.com firstname.lastname@example.org's password: Last login: Fri Feb 9 00:03:38 2007 from wsip-70-168-187-166.ga.at.cox.net [test@submit ~]$
ssh email@example.com is what you would type in at a command prompt on your system. After this, it asks you for a password, which you type in. After that, you are logged in and ready to work.
For Microsoft Windows, things are a bit trickier. Windows does not come with a built-in SSH client, and this makes things difficult. What you have to do is download a client from the Internet and install it, then use that client. We recommend the following:
Both of the above clients have documentation at their websites, so I will not go into it here. Once you are logged in and have a prompt that resembles that in the Unix / Linux section above, you can continue with this tutorial.
For Macintosh users that are running OS X (those who are not running OS X are in a whole different category of user) the instructions are very similar to those of Linux.
One thing that may be tough for a Macintosh user is finding a way in which to get to a command prompt in the first place so that they can use the linux-style commands. In order to get a command prompt, you need to start a terminal session. The terminal program on a Macintosh is located in Applications/Accessories. When a user starts this program, they start a terminal session whose prompt is located at that user's home directory.
From there, you should be able to follow the linux based instructions listed above.
If there is anyone out there that knows of a GUI application for Mac that makes the use of SCP easier, please let me know and I will include it here.
Changing your Password
The passwords assigned to users at the HPC Center expire every six months. If you receive a message stating that it is about to expire, you can reset it. In order to change your password to something perhaps a little easier to remember than the password supplied, which typically looks like a line of static, you can do the following:
$ passwd Changing password for user <YOUR USERNAME>. Enter login(LDAP) password: New UNIX password: Retype new UNIX password: New password: Re-enter new password: LDAP password information changed for <YOUR USERNAME> passwd: all authentication tokens updated successfully.
The above asks for your old password, then asks for your new password a total of four times. This seems like a lot, but it is actually a nice thing because it does the following:
- Makes sure that you are entering your new password correctly.
- Does a bit of training your fingers for muscle memory.
If your HPC password expires, you can reset your password here on the HPC website using your Gatorlink username and password.
We expect the users of the HPC Center cluster to already have a working knowledge of the linux operating system, so we are not going to go into detail here on using the operating system. Below are some links to webpages that describe a lot of this, however:
We have a structured file system that is important to know about. Please read about it here: HPC File System
Editing files on the cluster can be done through a couple of different methods...
- vi - vi is the basic editor for a number of people. Using the editor is not necessarily intuitive, so we have provided a link to a tutorial.
- emacs - emacs is a much heavier duty editor, but again has the problem of having commands that are non-intuitive. Again, we have provided a link to a tutorial for this editor.
- pico - While pico is not installed on the system, nano is installed, and is a clone of pico.
You can also use your favorite editor on your local machine and then transfer the files over to the HPC Cluster afterwards. One caveat to this is that with files created on Windows machines, very often extra characters get injected into text files which can cause major problems when it comes time to interpret the files. If this happens, there is a utility called
dos2unix that you can use to remove the extra characters.
Jobs from faculty investors in the HPC Center are now favored over jobs from groups who did not invest in the HPC Center.
Job scheduling has been a big topic with the HPC committee in the last several months. The HPC Center staff has been directed by the committee to improve the quality of service of job scheduling for jobs coming from investors in the HPC Center. This means reducing the time spent in the queues and allowing jobs from the investors to capture the full share of the resources that they have paid for. The HPC committee recently adopted a document which spells out what they want.
Jobs can be submitted on submit.hpc.ufl.edu.
The Torque Resource Manager has been installed on the HPC cluster, and is slowly being switched over to from the PBS Pro scheduler as time goes on. The Maui scheduler is also running in this environment.
The Torque scheduler is installed on the cluster, and accepts the same commands as the PBS Pro scheduler with a couple of exceptions. Currently we are experimenting with these packages so that we can provide improved scheduling to HPC users. While we are still learning about Torque and Maui, our experiences so far have been good and we are guardedly optimistic that Torque and Maui will end up being the resource manager and scheduler for the HPC Center sometime in the not-too-distant future.
Please note the following.
- If your job is single-threaded (1 cpu) and does not have heavy I/O requirements, it does not need to run on an infiniband-enabled node. In that case, you should include the "gige" property in your PBS resource specification as follows
#PBS -l nodes=1:ppn=1
- If you need to run an mpi-based application that has not been rebuilt for OpenMPI 1.2.0+Torque, please send us a note and we'll be happy to rebuild what you need - first come, first serve.
- If you build your own MPI-based application executables, you should use the MPI compiler wrappers (mpif90, mpicc, mpiCC) in /opt/intel/ompi/1.2.0/bin. These wrappers will automatically pull in the correct libraries.
- We will continue to tune the maui scheduler to provide fair and efficient scheduling according to the policies established by the HPC Committee and within the capabilities of the maui scheduler. Keep in mind that these policies include priority and quality-of-service commitments to those faculty who have invested in the resources within the HPC Center.
#! /bin/sh #PBS -N testjob #PBS -o testjob.out #PBS -e testjob.err #PBS -M <INSERT EMAIL HERE> #PBS -r n #PBS -l walltime=00:01:00 #PBS -l nodes=1:ppn=1 #PBS -l pmem=100mb date hostname
To submit this job from submit.hpc.ufl.edu, you would use the following command:
$ qsub <your job script>
To check the status of running jobs, you would use the following command:
$ qstat [-u <username>]
More information on PBS scripts can be seen with our Sample Scripts.
Notes on Batch Scripts
- The script can handle only one set of directives. Do not submit a script that has more than one set of directives included in it, as this will cause the Moab/Torque system to reject the script with a qsub: Job rejected by all possible destinations error. This problem was first seen when a user complained about a script that was being rejected with this error. Upon further inspection of their script, it was found that the script had concatenated versions of itself in the same file.
- For more info on advanced directives see PBS_Directives
Compiling your own
By default, when you first login to the system you have compilers setup for the Intel OpenMPI compilers. This gives you access to C, C++, F77 and F90 compilers in your path, as well as the mpicc and mpif90 compilers for OpenMPI applications. If you want to change this, you can use a selector program on the cluster called "mpi-selector". The command specification for this is as follows:
/usr/bin/mpi-selector options: Options for MPI implementations: --register <name> Register an MPI implementation with the central mpi-selector database. Requires use of the --source-dir option. --source-dir <dir> Used with --register, indicating that <dir> is where mpivars.sh and mpivars.csh can be found. --unregister <name> Remove an MPI implementation list from the central mpi-selector database. Options for system adminsitrators: --system When used with the --set and --unset options, act on the system-wide defaults (vs. the per-user defaults). When used with --query, only show the site-wide default (if there is one). --user When used with the --set and --unset options, act on the per-user defaults (vs. the site-wide defaults). When used with --query, only show the per-user default (if there is one). Options for users: --list Show a list of the currently registered MPI implementations. --set <name> Set <name> to be the default MPI selection. --unset Remove the default MPI selection. --query Shows the current default MPI selection. --yes Assume the answer is "yes" to any question. --no Assume the answer is "no" to any question. --verbose Be verbose about actions. --silent Print nothing (not even warnings or errors; overrides --verbose) --version Display the version of /usr/bin/mpi-selector.