Bigdog Linux Beowulf Cluster
Here is some useful information about the bigdog cluster
The College of PAMS has a Beowulf cluster for high performance computing. These webpages contain information about the cluster and support issues.
This cluster can be used for jobs that are parallel, that can be broken into many smaller parts, or long uninterrupted compute jobs. You will need be familiar with the Sun Grid Engine and the C-Shell for scripting. Also MPI programming is definately a plus for parallel jobs.
Beowulf Course: Using the ClusterPCO aims to have a course to teach users of the Beowulf how to best use the system twice a semester. Once at the beginning and once toward the middle of the semester. This course aims to familiarize users to our Beowulf environment, teach the usage of the batch processing system (the Sun Grid Engine), running MPI jobs, and pointers to how to get started programming with MPI. Lots of examples, source code, and resources for more information are included. Information on the next Beowulf Course will be emailed to those that currently have accounts on the Beowulf, posted to Sysnews, and sent out to the department heads in PAMS. Login informationThe head node of the cluster can be accessed via SSH at bigdog.pams.ncsu.edu. After you have been granted an account you should be able to log in with your NCSU username and password. Files should be transferred to and from the head node via SCP. All jobs must be submitted to the batch system with qsub or the cluster will not run your job. If you would like to apply for an account on the PAMS Beowulf please read about getting an account. Monitor Your JobsFor best performance you need to pay close attention to your jobs that are running on the Beowulf. This can help you figure out if your job is properly parallelized and if its run with the right options. For example, if a node that's running one of your processes has a load average that is greater than the number of CPUs in that node there may be a problem. Also, if a node running your job starts using a lot of swap space then your job is using too much memory and will run very slowly. The Cluster Status Page gives you all sorts of information about the cluster. Currently running jobs, memory usage, network usage, CPU usage, and many other metrics. Specifications on the Cluster
The cluster uses Red Hat Enterprise Linux version 3. The cluster consists of two main parts. The head nodes (AKA the login nodes) and the compute nodes. The head node is the computer that you log into via SSH and do your work from. It is responsible for distributing the work to each of the compute nodes. The cluster has 600 GB of RAID-5 storage for home directories, and 1 TB in /share where users can store temporary data. There are also several terrabytes available for data storage on an Apple XRaid. Access to the head node is available via SSH. The total computational power of the cluster is greater than 300 Gflops. Network Diagram of Beowulf Cluster. Third Party Software on the ClusterThe following is a list of 3rd party software that has been installed on the PAMS Beowulf. All software is mounted under /opt on each node. If you would like to request additional software to be installed please email us at pams_hpc@help.ncsu.edu . Note that with the new revison of the cluster you now have to "add" packages, as you do on normal realm linux machines.
|