CAEN

The Center for Advanced Computing (CAC) at the College of Engineering provides cluster-based high performance computing for many faculty, instructors, and researchers at the University of Michigan. This level of computing is accomplished via four computing clusters.

The cluster known as Morpheus consists of 104 nodes. Each node in this cluster has two AMD Athlon CPUs, and is running RedHat Linux. This system has an aggregate 200GB of RAM and 4.5TB of scratch disk space. The nodes in this cluster are interconnected via a high speed (2Gbps), low latency network from Myricom. This system is ideal for parallel programs that can benefit from fast communications.

The second Linux-based cluster includes 212 nodes and is known as Nyx. Each node in this cluster has two AMD Opteron CPUs, and is running SuSE Linux. Nyx has an aggregate 580GB of RAM and nearly 8TB of scratch disk space. The nodes in this cluster are interconnected via a high speed (1Gbps) Ethernet switch from Force-10. This system is ideal for programs that can benefit from a 64-bit processor, or for serial programs.

CAC also has a 64-node Apple cluster. Each node has two G5 processors, and is running MacOS X. This cluster has an aggregate 128GB of RAM and more than 4.5TB of scratch disk space. The nodes in this cluster are interconnected using the same Force-10 gigabit Ethernet switch mentioned above.

CAC also maintains a 160-node Apple cluster that was purchased by faculty in the College of Engineering. This system includes dual G5 processor nodes running MacOS X, and has an aggregate 160GB of RAM and nearly 11TB of scratch disk space. The network interconnects for these nodes are standard 100Mb Ethernet because the type of computation done with this cluster does not require high-performance communication between the parallel processes. Usage

These large clusters are used by researchers to solve a diverse array of science and engineering problems. Some of the types of computations being done include:

• Monte Carlo simulation of radioimmunotherapy treatment
• Computational materials science for nano-materials
• Predictive modeling of space-weather events
• Aeroelasticity calculations for aircraft that travel at hypersonic speeds
• Computational modeling of the human heart
• Parallel video rendering
• Collaborative data processing of information from the large hadron collider at the CERN laboratory

This large spectrum of research activities comes from all over the University. CAC makes every attempt to support researchers from both the College of Engineering and from the other colleges on campus.

For researchers and students working on problems that can be solved using parallel computing methods, CAC encourages the use of the MPI parallel library. This freely-available library is used by many commercial, open-source, and private parallel programs.

If your research does not lend itself to being parallelized, you may still find the CAC resources of interest. CAC offers systems with larger-than-usual memory configurations (up to 6GB per node today with plans for 16GB per node for some nodes) that support very long-running processes on the order of weeks of wall-clock time. Access to the cluster nodes is guaranteed to be exclusive. If your job is running, it does not share the resources of that node with any other user.

Job scheduling on CAC-maintained systems is managed by a batch queuing system (currently PBSPro). Each job is submitted to the queue and based on its needs (memory, CPU speed, run time, etc.) it is started as soon as possible. On the largest cluster, the average wait time for a job to start is about four hours. CAC faculty and staff

Kenneth G. Powell, Professor of Aerospace Engineering, is the director of CAC. CAC staff, consisting of four full-time employees and a student employee, work closely with CAEN. Systems support is done by Matthew Britt, Andrew Caird, and Brock Palen. Matt has been with CAC for six years; Andy recently returned to the CAC after seven years of working in industry; and Brock, a Nuclear Engineering student, has been with CAC for nearly a year. Programming and application support is done by Dr. David Woodcock and Dr. Abhijit Bose. Dr. Woodcock has been with CAC and its predecessors (CPC and LaSC) for more than fifteen years and Dr. Bose has been with CAC for four years. Support and access to CAC

U-M researchers may gain access to these clusters. If you are interested, contact Professor Powell by sending email to powell@umich.edu. CAC staff occasionally offers short courses on beginning parallel programming and general system usage, and may provide individual consulting as well. Contact CAC staff by email to cac-info@umich.edu. Additional information can be found on the CAC web site at cac.engin.umich.edu.