Dr. Peter Sobe, research assistant at the University, told NDCF that the cluster, named Magellan after the 16th Century explorer, was installed in spring 2004. It now contains eight server blades from Angstrom Microsystems, each with dual Advanced Micro Devices (NYSE: AMD) Opteron processors, and a total disk capacity of 640 Gigabytes.
The cluster, which is located within the Universitys Institute of Computer Engineering, is capable of a peak performance of 98 million operations per second, according to Sobe. Terragrid software from Montreal, Canada-based startup Terrascale Technologies Inc. provides access to I/O capacity across the cluster.
The software works with an InfiniBand switch from Topspin Communications Inc. to turn the network into what looks like a global disk drive to which all the machines can read and write, according to Gautham Sastri, Terrascales CEO.
Though Gigabit Ethernet is also deployed within the cluster, InfiniBand is the main transport. At the time of purchase [InfiniBand] was the best choice in transfer capacity, price, and availability, Sobe says.