The High Performance Computing (HPC) group of Jefferson Lab is conducting research and development in parallel cluster computing, parallel job scheduling and high performance messaging software to support a national effort in terascale simulations of Quantum Chromodynamics (QCD). Lattice QCD, a numerical approach to QCD, is used to calculate weak decays of strongly interacting particles, to investigate matter under extreme conditions and to study the structure and /interaction of hadrons.
The HPC group currently manages three computing clusters:
 |
A 396 nodes InfiniBand cluster. Each node contains two AMD Quad-Core Opteron Processors and a PCI-Express (x16) slot for an InfiniBand HCA Adapter that provides 20Gb/s network bandwidth. This cluster delivers more than 13.46 TFlops performance in terms of the Linpack benchmark. |
 |
A 280 dual-core processor InfiniBand cluster based on Intel 3.0 GHz Pentium D (Dual Core) CPUs and 800 MHz front side bus. Each node has a PCI-Express (x4) slot for an InfiniBand HCA Adapter that provides 10Gb/s network bandwidth. |
 |
A 384 processor 3D gigabit ethernet mesh cluster based on 2.8 GHz Intel Pentium 4 Xeon CPUs and 800 MHz front side bus. Interesting feature: no high performance switch; each node is directly wired to 7 adjacent nodes. Lower cost per node, yet enormous bandwidth, limited only by memory and PCI-X bandwidths. |
The HPC group is also participating in the Particle Physics Data Grid project to develop location independent access to replicated data sets. This work will support both access to Jefferson Lab experimental physics data, and access to LQCD calculation data.
Staff members of the HPC group are also involved in other areas of
software development such as the distributed control middleware and distributed
message logging system which are used in operations in both
national
and international laboratories.