HPC | Computer Science and Engineering | University of Arkansas Home Page

Supercomputing
@ University of Arkansas

With teraflops of available computer power at their disposal, researchers at the University of Arkansas are solving more computationally intensive problems than ever before.

Star of Arkansas — Arkansas's Newest Supercomputing Cluster














The Star of Arkansas is funded under Grant MRI #0722625 "Acquisition of a Supercomputing Cluster for Computational and Data-Intensive Applications in Science and Engineering" from the National Science Foundation, with substantial matching funds from the University of Arkansas, and in partnership with Dell Corporation.

The Star of Arkansas supercomputer is built from 157 compute nodes, each with dual quad-core Xeon E5430 processors, 2x6MB cache, 2.66GHz, 1333FSB. There are a total of 1,256 cores, and each core has 2GB of memory. Performance on supercomputers is measured in "flops," or floating point operations per second. The theoretical peak performance of Star is 13.36Tflops, or 13.36 trillion floating point operations each second.

The Star of Arkansas supercomputer is interconnected with an InfiniBand network that runs at 10Gbps, or 10 billions bits of information every second. The IB switch is from Qlogic and has expansion capacity up to 256 slots, of which only 166 are used in the current configuration. With this switch additional compute nodes can be purchased incrementally and added to the cluster as additional hardware funds become available. The cluster is interconnected with an additional Gigabit Ethernet network for NFS access, and another Gigabit Ethernet network for management.

The Star of Arkansas has two file systems to handle computational and data-intensive applications. The NFS file system will be used for permanent storage and consists of 4TB of raw disk storage, or more than 4 trillion characters of data. The Lustre file system is used for fast temporary storage. The storage for Lustre comes from Data Direct Networks and includes 21 TB of raw disk storage. In addition, 50 nodes have local storage of 1TB, and 107 nodes have local storage of about 320GB, more than 320 billion characters of data. The total amount of storage potentially available on Star is more than 109TB. In comparison, the Library of Congress has estimated that it stores 20TB of text.


Red Diamond — UofA's 1st Supercomputing Cluster

Red Diamond is a cluster of 128 dual-processor compute nodes utilizing the 64-bit Intel® Xeon™ processor. Computers are interconnected with an InfiniBand high-speed network, supplied by Topspin, which provides data throughput at over 800 megabytes per second and a latency of less than six microseconds. Dell™ is partnering with the University of Arkansas as the supplier and integrator of Red Diamond.

Red Diamond is the first supercomputer in Arkansas. Its more than ten terabytes of external storage and processing capabilities will allow research results from the University of Arkansas to be accessible to researchers and to be calculated by researchers on the Internet worldwide.

Funding for Red Diamond is supplied through a Major Research Instrumentation Grant (#0421099) from the National Science Foundation and additional support from the University of Arkansas.

Why "Red Diamond"? The name was chosen as a symbol of the first supercomputer in the state of Arkansas for new discoveries in science and engineering. Arkansas is the home of the world's only public diamond mine and the eighth largest diamond repository in the world. The largest diamond ever found in the United States, originally weighing 40.23 carats, was found in Arkansas. The color red comes from the school colors of the University of Arkansas and from the colors of the Arkansas state flag, which boasts a diamond on a red field to remember the state as a place where diamonds are discovered and mined.


Trillion — Mechanical Engineering Cluster

Trillion is a 96 CPU cluster with AMD Opteron 252 processors and 4 Gb of RAM per node. Computers are interconnected using split Gig-E communication in which MPI traffic is isolated to a dedicated internal network. Trillion delivers an Rmax of 336 Gflops and is owned by the Department of Mechanical Engineering.

Last updated: February 01, 2008