Berlin 2005 – wissenschaftliches Programm

Bereiche | Tage | Auswahl | Suche | Downloads | Hilfe

T: Teilchenphysik

T 407: Computing I

T 407.1: Vortrag

Montag, 7. März 2005, 16:30–16:45, TU H3002

Reliable Mass Storage on Unreliable Components — •Arne Wiebalck, Sebastian Kalcher, and Volker Lindenstruth — KIP, University Heidelberg

Extrapolating the current trends in hard drive technology, the PC clusters deployed by the upcoming high energy physics experiments will provide an online disk capacity in the range of up to 1 petabyte. This amount is significant when being compared to the requirements of the individual experiments for mass storage. The reason why this capacity is often not used, for instance as a complement for more traditional tape-based systems, is its inherent unreliablity. The talk will set out the general architecture and prototype results of the ClusterRAID, a reliable, distributed mass storage system for clusters. The key paradigm of this system is to make the local hard drive a reliable device and thus it reflects two (often neglected) aspects common to many applications, and to high energy physics data processing in particular: the locality of disk accesses and the independence of tasks. The system provides adjustable reliability as it is capable of tolerating multiple simultaneous disk or node failures. Nonetheless, only the theoretical minimum of network transfers is required due to the use of sophisticated error-correcting codes.