search query: @supervisor Nurminen, Jukka / total: 45
reference: 2 / 45
| Author: | Pestana, Goncalo Marques |
| Title: | Energy Efficiency in High Throughput Computing Tools, techniques and experiments |
| Publication type: | Master's thesis |
| Publication year: | 2016 |
| Pages: | 72 Language: eng |
| Department/School: | Perustieteiden korkeakoulu |
| Main subject: | (T3005) |
| Supervisor: | Nurminen, Jukka |
| Instructor: | Ou, Zhonghong ; Niemi, Tapio |
| Electronic version URL: | http://urn.fi/URN:NBN:fi:aalto-201610124915 |
| Location: | P1 Ark Aalto 5684 | Archive |
| Keywords: | energy efficiency scientific computing ARM Intel RAPL |
| Abstract (eng): | The volume of data to process and store in high throughput computing (HTC) and scientific computing continues increasing many-fold every year. Consequently, the energy consumption of data centers and similar facilities is raising economical and environmental concerns. Thus, it is of paramount importance to improve energy efficiency in such environments. This thesis focuses on understanding how to improve energy efficiency in scientific computing and HTC. For this purpose we conducted research on tools and techniques to measure power consumption. We also conducted experiments to understand if low-energy processing architectures are suitable for HTC and compared the energy efficiency of ARM and Intel architectures under authentic scientific workloads. Finally, we used the results to develop an algorithm that schedules tasks among ARM and Intel machines in a dynamic electricity pricing market in order to optimally lower the overall electricity bill. Our contributions are three-fold: The results of the study indicate that ARM has potential for being used in scientific and HTC from an energy efficiency perspective; We also outlined a set of tools and techniques to accurately measure energy consumption at the different levels of the computing systems; In addiciton, the developed scheduling algorithm shows potential savings in the electrical bill when applied to heterogeneous data centers working under a dynamic electricity pricing market. |
| ED: | 2016-10-16 |
INSSI record number: 54647
+ add basket
INSSI