Using Multitasking and SSD Disks for Optimising Computing Cluster Energy-Efficiency

Tapio Niemi, Jukka Kommeri, Ari-Pekka Hameri, Hec
We tested how solid-state drives (SSD) and multi-tasking, i.e., running more than one task per CPU core, affect performance and energy consumption in I/O intensive data analysis jobs in high-energy physics (HEP) computing. Our motivation for the study comes from the LHC experiment at CERN producing up to 15 petabytes of data annually. The data is analysed by a huge grid computing infrastructure containing more than 100 000 CPU cores over 140 computing centres in different countries. Our tests
more » ... ntries. Our tests indicated that in I/O intensive HEP computing multitasking, mixing heterogeneous tasks, and using SSD disks can clearly improve throughput and lower energy consumption. The throughput improved around 120% and energy consumption decreased around 50% compared to a conventional one task per CPU core setting with a hard disk system.