|
 The Theoretical
Astrophysics Group at Northwestern University
maintains a beowulf-cluster, named Fugu, for its
computing needs. Fugu is utilized by professors,
post-doctorate researchers, graduate students,
and undergraduates for
theoretical calculations including extrasolar
planet dynamics, galactic and globular cluster
dynamics, stellar evolution, and compact
object formation.
Partially funded by an NSF MRI award earned by Professor Kalogera, Fugu has functionally replaced the group's old computing cluster, Typhoon. Fugu is composed of 78 general-purpose nodes and two dedicated nodes designed specifically for n-body calculations, as well as a large compute node optimized for parallel tasks. Each node contains either two quad core 2.2 GHz (30) or two dual core 2.8 GHz (48) AMD Opteron processors with 2 GB of RAM per core. The large compute node features 4 quad core 2.2 GHz AMD Opteron processors and 32 GB of RAM as well as four 146GB SAS 15k RPM hard drives. There are 13 TB of hard disk space available in a RAID-6 array which is attached to the master node via mini SAS channel. The master node is connected to the central network switch via a 10 Gb/s Ethernet adapter, while the rest of the nodes are connected using a 1 Gb/s network. In total, Fugu's 448 cores have a theoretical peak performance of 3.38 TFLOPS.
Fugu runs on Linux Rocks 5 and makes use of the Sun Grid Engine scheduler, allowing both serial and parallel jobs as well as checkpointing.
In September 2008, Fugu will be expanded with 10 GPU-Compute nodes to take
advantage of the extremely parallel computation power of GPUs in HPC
applications. The expansion is funded partly by the NSF MRI award and
partly by Academic & Research Technologies at NU. Each node will contain 1
quad core AMD Phenom 9950 Black Edition CPU (2.6 GHz) with 2 GB of RAM per
core and 3 Nvidia GeForce GTX 280 graphics cards with 1GB of 512-bit GDDR3
memory. These GPUs will provide Fugu with a peak performance of about 30 TFLOPS.
|
