Parts | Days | Selection | Search | Downloads | Help
HK: Physik der Hadronen und Kerne
HK 41: Instrumentation und Anwendungen
HK 41.4: Talk
Wednesday, March 9, 2005, 14:45–15:00, TU MA041
Prototype of a Dedicated Multi-Node Data Processing System for Realtime Trigger and Analysis Applications — •Daniel Kirschner, Marco Destefanis, Ingo Fröhlich, Camilla Gilardi, Wolfgang Kühn, Francesca Ottone, Vladimir Pechenov, and Tiago Perez — II. Phys. Inst. Giessen, Heinrich-Buff-Ring 14, 35392 Giessen
Modern experiments in hadron physics like the HADES detector at GSI-Darmstadt produce a large amount of data that has to be distributed, stored and analyzed. Analysis of this data is very time consuming due to the large amount of data and the complex algorithms needed. This problem can be addressed by a dedicated multi-node and multi-CPU computing architecture interconnected by Gigabit-Ethernet. Dedicated hardware has the advantages over “Grid-Computers” in skaleability, price per computational unit, predictability of time behavior (possibility of real time applications) and ease of administration. Gigabit-Ethernet provides an efficient and standardized infrastructure for data distribution. This infrastructure can be used to distribute data in an experiment as well as to distribute data in a multi-node computing environment. The prototype VME-Bus card has of two major units: a network unit featuring two Gigabit Ethernet over Copper connections and a computational part featuring a TigerSHARC Digital Signal Processor. The discussion will concentrate the real-life performance of Gigabit Ethernet as the main topic. Supported by: BMBF 06 GI 144, BMBF 06 GI 145.