Wuppertal 2015 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
T: Fachverband Teilchenphysik
T 76: Grid-Computing I
T 76.4: Vortrag
Mittwoch, 11. März 2015, 17:30–17:45, K.11.20 (K5)
Hadoop as cluster file system for an ATLAS Tier3 analysis site — •Rui Zhang, Philip Bechtle, Ian C. Brock, and Jan A. Stillings for the ATLAS collaboration — University of Bonn, Germany
The Apache Hadoop Distributed File System (HDFSTM) is widely used in cloud computing, aiming at minimising the impact of finite network bandwidth between computing and storage elements. The computing happens on the node where the data is held, avoiding the distinction between storage and computing elements. For high I/O demands as in HEP, this is of interest, since the limited network bandwidth can present a bottleneck for the efficiency of the whole cluster.
The performance of a Hadoop system is tested on a prototype. An ATLAS example analysis is ported following a MapReduce programming model. Discussions about the feasibility and potential in HEP are presented. The cost of the migration to a Hadoop system for a typical user is evaluated.