Our group has been active in the area of parallel computing and commodity parallel computing since the earliest days. We were involved in the original Beowulf project and built the first one outside of NASA Goddard Space Flight Center and the very first one to use CPUs other than Intel. Four generations of zBox computers have provided a local experimental platform. It has also let our students experience building supercomputers.
Our in-house designed (Joachim Stadel & Ben Moore 2003), massively parallel supercomputer for running our cosmological N-body simulations.
The mass storage system 'smaug' was installed in late 2006. It provided unlimited capacity on tape (~100 TB), while holding only the most active data online in a ~30 TB disk cache with 150 disks.
The zBox2 was installed in January 2006 with 125 quad opteron 852 nodes and a total of 580Gb ram and 65 Tb of disk. The nodes are connected with a high speed 3D SCI network.
The zBox in its frame.
The fourth generation of self-made supercomputer at the ICS represents the first real design update to the original zBox1 which was built in 2002. The number of cores and amount of memory was increased substantially over zBox3 and the very aging SCI Network replaced by QDR Infiniband. From the 576 cores (Intel Core2) and 1.3 TB RAM of zBox3, we now have 3072 cores (Intel Sandy Bridge E5) and 12 TB of RAM. The zBox4 platters were completely redesigned and now hold 4 nodes each. We also added special nozzle to improve airflow and cleaning up the way cables are routed. The zBox4 construction began end of September 2012 and was performed by volunteer students, postdocs and friends of the Institute, with even a few professors lending a helping hand.
We currently have 1PB of data on spinning disk and have moved as much data from a single simulation to the Johns Hopkins Datascope for visualization and analysis.
More information available at the dedicated zBox4 webpage.