We've all been there. You know the moment: sitting at your computer just wondering when it would finish the current task so you could move on with your life. If you have a new computer, this probably does not happen to you very often. That is, of course, unless you wish to run complex mathematical operations or simulations on your computer. To effectively run such software, one must rely upon the power of super computers.
Simply put, a super computer is a computer with access to vast amounts of resources. This could mean more processors, faster processor speed, more memory, or larger hard drive space. Here at NC State, there are three main high performance computing (HPC) facilities. They are the ITD's High Performance Center, LSF, the PAMS Beowulf cluster, and Wolfgrid.
In July 2003, The NC State Information Technology Division put into production an IBM p690 acquired from the now defunct North Carolina Supercomputing Center. The system harnesses the power of 32 Power4 1.3 Ghz processors and has 128GB of shared memory, with access to over 10TB of storage space.
New to NCSU is the IBM Blade Center #1; it makes use of 32 Xeon 2.4 Ghz processors and 48GB of distributed memory, meaning each processor has 3GB. It has access to 1.2TB of hard drive space with a 1Gbps connection to mass storage. For more information on running jobs on this computer, please visit the Getting Started with the IBM Blade Center page.
A third super computer is being assembled. In all, it will consist of 64 Xeon 2.8Ghz processors. This computer will have 256Gb of distributed memory. There will be a 2.56Tb hard drive for storage. And, as soon as possible, Myrinet and a 1Gbps connection with a larger mass storage device.
For information on accessing these computers, visit ITD's High Performance Computer page. Current software packages already installed or licensed include:
- Amber 7 - A suite of molecular simulation programs
- Ansys - A design simulation program for structural/physical analysis
- ARPS - Atmospheric modeling and prediction software
- PAUP - Software used to track and make inferences on evolutionary trees
- NCAR - A utility for drawing various maps and diagrams
The Condor Project is currently under way within the ECE department. This is an effort to bring the Condor software to NC State. Condor software provides a specialized workload management system for a network of computers, allowing for the unused clock cycles of computers on a network to be exploited.
When a computer goes idle, the Condor software will recognize this and look to its queue to see if there are any jobs waiting to be run. If it finds jobs, they are run, while the computer stays idle. When, however, someone comes to and uses the computer, Condor instructs itself to set a checkpoint and stops processing the data.
Condor at NC State is still in developmental stages, but hopefully will be available soon. Both a Windows and Red Hat Linux solution are being developed. More information will be made available as soon as the project nears completion.
Need info on LSF @ NCState.
CSC ARC Cluster
1728 cores on 108 compute nodes integrated by Advanced HPC. All machines are 2-way SMPs with AMD Opteron 6128 (Magny Core) processors with 8 cores per socket (16 cores per node). For details, including how to get access: http://moss.csc.ncsu.edu/~mueller/cluster/arc/
PAMS Beowulf Cluster
The College of PAMS, about three years ago, purchased thirty computers to be used in a Beowulf class cluster. Today, that number has grown the include 87 computing nodes, containing 213 CPUs. These facilities are intended for the aid of students and faculty in the College of PAMS, but have been extended to the entire NCSU community.
The Beowulf cluster at NC State is mainly to be used for parallel jobs--jobs that can be broken down into small units. Like all Beowulf computers, the College of PAMS Beowulf is isolated from the rest of the world. You are only allowed access the head node, which then breaks apart and distributes your jobs across the network. This enables the compute nodes to focus only on processing jobs and not the presentation. This allows for quicker processing of jobs and better end results.
The cluster consists of two parts, the head node and the compute nodes. You SSH into the head node and it distributes the work to the compute nodes. There is 135 GB or RADI-5 storage and 1 TB in /tmpspace for user storage and temporary data. The total computational power to this Beowulf cluster is in excess of 300 Gflops. For more information, please see the PAMS High Performance Computing webpage.
Wolfgrid *was* a community supercomputer that allows any recent apple computer to connect to Wolfgrid and will then process jobs that you submit to it. You can also become part of Wolfgrid by letting your computer become a processing node (an agent). You can set your computer to only act as an agent when the screen saver comes on so it will not disrupt your use of the computer. Wolfgrid puts the power in your hands, where you have a supercomputer grid working for you.
This particular project seems to have fallen by the wayside. Though if there is interest, we may be able to revive it. We're investigating the potential requirements, investment, and interest from the academic/research community. Any comments are welcome.