Mainz 2014 – wissenschaftliches Programm

Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

T: Fachverband Teilchenphysik

T 12: Computing

T 12.1: Vortrag

Montag, 24. März 2014, 11:00–11:15, P15

High Performance Computing usage for ATLAS — •Rodney Walker — Ludwig-Maximilians-Universitaet, Muenchen

Current ATLAS computing is largely provided by a distributed federation of High Throughput Computing(HTC) clusters, based at universities and labs around the world. The model was well grounded and critical to the success of LHC Run 1 data analysis, but budgetary constraints mean that it is unlikely to be sufficient for Run 2 and beyond.

In the meantime, High Performance Computing(HPC) has evolved to become dominated by large x86 linux clusters which, despite the specialized hardware, are comparable in cost to the smaller university-based HTC linux clusters. Indeed, economies of scale, together with novel energy saving features, can make the cost of ownership significantly cheaper. Such HPC clusters are well suited for standard ATLAS production work, although there are a number of hurdles to overcome - technical solutions and working examples will be described.

We also discuss specific ATLAS workflows which can take advantage of the HPC capabilities, such as fast interconnect and high performance files systems. Allocated cpu-time on such general scientific resources, together with the large potential for opportunistic use, by means of backfill, should provide a valuable contribution to ATLAS's future computing needs.