marylou4 Frequently Asked QuestionsHelpQ: Who are the system managers and how do I contact them?
A: Any of the systems administrators can assist you with marylou4. Please visit our staff page for contact information.Q: What other help resources are available?
A: This FAQ addresses many marylou4-specific technical issues; if you can't find the answer you're looking for here, either contact the system administration staff or open a case in our ticket tracker.System ConfigurationQ: What is marylou4's hardware configuration?
A: marylou4 is a Dell 1955 Linux cluster. There are 630 nodes, each with two Intel 2.6GHz Xeon EM64T dual-core processors and 8 GB of fully-buffered RAM. 210 of the nodes are connected with an Infiniband switch for MPI performance, as well as standard Ethernet. All 630 nodes are connected with Gigabit Ethernet.Q: What is marylou4's software configuration?
A: marylou4 runs the Linux kernel (Red Hat Enterprise Linux v4). The cluster's EM64T processors run a 64-bit Linux kernel. Most standard Linux applications and utilities are available, including the GNU compilers and utilities. In addition, optimized compilers from Intel are available.Accounts and AccessQ: How do I request an account?
A: Follow the Getting an Account link in the Main Menu navigation pane on the left side of this page. Your application will usually be processed within 2 business days and you will receive an email notification when your account is approved. Student accounts require a faculty mentor/sponsor to also have an account.Q: Why do faculty have to have an account in order for their students to have an account?
A: Faculty are required to have an account for these and other reasons: 1- It aids in generating FSL use statistics, 2- Creates a head to the research group for fairshare and job priority calculations, 3- To help verify annually the legitimacy of student accounts under their sponsorship, and 4- To annually provide FSL with feedback on effectiveness of the system, and report # of publications and monies received attributable to having access to FSL systems.Q: How do I connect to marylou4?
A: Access to marylou4 is available via the SSH protocol (version 2). Connect to the hostname marylou4.byu.edu. (Windows SSH2 client software is available from The PuTTY download page and many other vendors; Linux and Apple OS X contain the SSH software as part of the operating system).Q: How do I change my password and/or login shell?
A: You can change your password and shell from the account management page. (Link in left navigation pane on all FSL web pages). Changes may take up to an hour to propagate.Q: How do I copy files to or from the supercomputers?
A: You can copy files using any program that can use the ssh protocol. The most common programs are scp and rsync. The main difference between rsync and scp is that rsync is designed to synchronize files and folders while scp is good for basic file copying. For more information, check out the man pages for both by typing 'man scp' or 'man rsync' in a Linux terminal. As a note, scp cannot distinguish between symbolic and hard links.SSH, Login and EnvironmentQ: Why do I get failure messages when I attempt to ssh/rsh/telnet between nodes?
A: This error most often occurs when you do not have SSH keys set up correctly. The first time you log in, the system should automatically prompt you for the appropriate information to generate those keys. When it prompts you for a passphrase, leave it blank. If you need to re-generate these keys, delete the following two files in your home directory, and then log in again: .ssh/id_rsa and .ssh/id_rsa.pub.Q: Why am I sent to a different node each time I open an SSH session to marylou4?
A: marylou4 uses load balancing software to manage the load for interactive sessions. When a user initiates an SSH session, the software routes the session to the interactive node with the lightest load. This means that you will likely be routed to a different node each time you open a session.Q: Why can't I access marylou4 from off of BYU campus?
A: You can. In the past, we did not allow this, but as of January 2007, you can connect from off-campus using marylou4.byu.eduQ: How do I configure my shell environment correctly for PBS, MPICH, etc?
A: By default, the system will automatically set up your environment correctly for most uses. However, it is possible to have the necessary variables reset by bash or tcsh profile files, like .bashrc and .cshrc in your home directory.Application DevelopmentQ: Which compilers are available?
A: The GNU and Intel compilers are available, for both serial and MPICH applications. The following table lists the commands available for each type of compiler:
Batch JobsQ: How do I submit and monitor batch jobs?
A: Please see our PBS Information and PBS Tutorial pages for this information. Please be aware, though, that the tutorial is designed for SMP systems like marylou, and not clusters like marylou4. The information page should describe the differences, however.Q: What are the limits on batch jobs?
A: marylou4's job limits are constantly adjusted to provide better usage patterns across the entire system. Generally extremely long jobs, extremely high numbers of jobs, or jobs requiring extremely high number of nodes are considered inconsiderate toward other users, and may be blocked by the system. If you run checkjob on your job, and it indicates a problem with your job, you can contact the system administrators for more information.Q: What is marylou4's scratch space configuration for batch jobs?
A: Each node has access to the high performance scratch volume on the Ibrix parallel filesystem, /fslhome/username/compute. There is a quota on this file system, and it is shared with user home directories, so please be careful how much space you occupy. By default that quota is 500 GB, and applies to all files owned by your user anywhere on the file system, both in your home directory, and in scratch. It is believed that this will not be restrictive for most users, but will keep a single individual from filling the entire file system. Those who abuse the filesystem will have their accounts disabled.Q: Can I run MPI jobs without using PBS?
A: Use of PBS is mandatory for all batch jobs submitted to the compute nodes. Use of the "mpirun" program to run jobs on the compute nodes (without using PBS) causes conflicts with jobs and reservations on the compute nodes, and so users who persist in doing so will have their accounts disabled.Q: Is there any way to test an MPI job before submitting it to Moab/PBS?
A: Sometimes it is desirable to test a program before submitting it to Moab and waiting several hours or days before it is run. To that end, we have created a special test queue that may be used. In addition, it is permissible to use the interactive nodes to test your MPI code. To do this you must use the -machinefile argument of the mpirun script. This argument points to a text file that lists permissible nodes, one per line. This text file should include only nodes m4int01 through m4int05. If you execute MPI jobs without using the -machinefile argument (or fail to specify the correct nodes in the machinefile), your account will be disabled. See the mpirun man page ('man mpirun' from the marylou4 shell) for more details on the -machinefile argument. Also, please remember that running mpirun manually on the interactive nodes is for testing only. Your MPI jobs on the interactive nodes should not run for more than a few minutes.Q: My MPI test job will not terminate or is running improperly on compute nodes. What should I do?
A: Sometimes MPI jobs (particularly those that are written as test jobs while learning to code in the MPI environment) do not terminate gracefully. If your job does not seem to be terminating properly (or if you realize you have accidentally started it on the compute nodes by mistake), you should immediately contact the system administrators. Explain your circumstances and exactly what you have done to the best of your recollection.Q: Why are there so many single-processor jobs? I thought this was a parallel cluster.
A: marylou4 is intended for use by a variety of users to provide them with computing resources unavailable to them elsewhere. In many cases this involves parallel message-passing code over the high-speed interconnect. In other cases it involves a single job or set of single-processor jobs running concurrently. All jobs are given equal preference. However there is a slight bias in favor of parallel jobs to offset the inherent flexibility of multiple single-processor submissions. We monitor the cluster usage carefully and adjust the scheduler to prevent one type of job from gaining dominance in the queue. |
CalendarNotices
|