Linux cluster
ARC has built and a high performance computing cluster for research projects which require parallel processing. Many complex computational problems are more efficient when broken down into small problems running in parallel. Examples are weather forecasting and fluid systems modelling, phylogentic tree building, and modelling moleculer interactions. Properly constructed software and the right problem can result in massive performance increase over the same software running in a symmetric multiprocessing system with a much lower hardware cost.
Our parallel cluster is constructed from a number of powerful servers capable of running in parallel and communicating via a gigabit ethernet network.
Rocks Cluster - 48 Cores/Queues status
- ROCKS 4.3 release using CentOS 4.1.
- Portland Group Compiler Suite with libraries optimized for the Opteron processor.
- Current cluster performance is measured at ~1 to ~9 Giga-FLOPS based on Linpack benchmarks.
- Cluster consists of the following nodes:
- 16 dual Opteron compute nodes, each with 4GB RAM and some ~30GB hd
- 10 IBM e325
- 6(5 + headnode) SUN V20Z
- Each node has 2 single-core AMD Opteron processors.
- Each node has 4GB of RAM.
- 4 Dual Opteron compute nodes, each with 4GB RAM and some ~30GB hd
- 4 SUN X2200 Dual-Dual Compute Nodes
- Each node has 2 dual-core AMD Opteron processors.
Gravel Cluster - 128 cores/queues status
- 23 Octo-Cores (2x4 Cores Intel Xeon) compute nodes - together 184 Cores
- each compute node has 8GB of ECC SDRAM
- The headnode is a Quad Core with 4 GB of SDRAM.
- ~2.5 TB of userspace Disk on local network
- Intel proprietary drivers and utilities Installed
- PGI compilers available - soon
XGrid Test Cluster - 8 cores/queues
- 4 2GHz Intel Core 2 Duo workstations
- 2 GB 667 Mhz DDR2 SDRAM per workstation
- Please contact us in case of further interested on this test cluster
ARC is working with PSU faculty to increase the number of nodes available for tackling large problems and maximizing our parallel computing capability. We are planning on adding additional nodes in the near future bringing the total number of compute nodes to 14.
The PSU cluster is open to students, faculty and developpers that are interested in developping, testing or using MPI enabled applications.
More technical information at the Documentation, Manuals and FAQ's for Rocks and Gravel Cluster(s) pages.