50 likes | 148 Views
Energy- efficient computing. How power constraints impact computing systems and applications ?. Ariel Oleksiak, Poznan Supercomputing and Networking Center (PSNC). ICT Energy consumption in numbers. 2 % of energy consumption globally Equal to aviation industry!
E N D
Energy-efficientcomputing How powerconstraintsimpactcomputingsystems and applications? Ariel Oleksiak, PoznanSupercomputing and Networking Center (PSNC)
ICT Energy consumptionin numbers • 2% of energy consumption globally • Equal to aviation industry! • Limits for building next generation exascalesupercomputers • Fastest supercomputers from TOP500 <1PFlops/MW! • “Greenest” supercomputers from Green500 ~2PFlops/MW • Limits for data centers • Bills reaching millions $ • Much energy for cooling and AC/DC conversion • Still often with PUE >>1 (e.g. 1.6-2.2 in 2010 study) • Rising energy costs: ~30% in the last decade in EU
Noveltrendsand challenges IT hardware • Move to massive parallelism using energy-efficient manycorearchitectures • From embedded devices market: many simple, low power cores (e.g. ARMs) • From highly specialized processors from gaming/graphics market space: GPU/Accelerator (e.g. NVidia Fermi, Tesla) • Massively parallel hybrid systems • Management of hybrid systems • Optimal mapping applications to hardware, application profiles • Applicationstaking advantage of hybrid systems • Hardware-software co-design, customized hardware • Modelingand simulations
Noveltrends and challengesInfrastructure • Huge infrastructures vsmicro-datacenters • Soon a single rack will deliver 1PFlop! • Wide distribution of resources/providers as envisioned in grids? • Novel approaches to managing energy supply • Use of renewable energy sources and batteries • Automated distribution, reacting to fluctuations • Use of direct current • Reducing conversion losses • Heat re-use • Autonomous data center management • Monitoring, measurements and metrics • Thermal-aware resource management, cooling system, energy supply