Loading...

Thermal-aware Accelerator Placement and Task Assignment for Energy Improvement in Data Center

Kazemi Abharian, Sanaz | 2016

365 Viewed
  1. Type of Document: M.Sc. Thesis
  2. Language: Farsi
  3. Document No: 49026 (19)
  4. University: Sharif University of Technology
  5. Department: Computer Engineering
  6. Advisor(s): Goudarzi, Maziar
  7. Abstract:
  8. With the proliferation of data centers, their ever-growing energy consumption has gained lots of attention from both academy and industry . Two primary parts that use majority of energy in data centers are IT equipment and cooling system or Computing Room Air Conditioning (CRAC) unit. Energy consumption of cooling system strongly relies on thermal performance of data center. Therefore, applying thermal management techniques for decreasing energy consumption of CRAC is a common practice. Moreover, the energy consumption of IT equipment affects the energy consumption of CRAC directly. Demand for more computing resources in data centers and their physical limits has, motivated the use of FPGAs as processing elements along with servers. Although FPGAs can add to computing capacity of data centers, they raise two concerns that need to be addressed. The first concern is the effect of FPGAs on the thermal performance of data center, and consequently the energy consumption of CRAC. The other concern is the energy consumption of FPGAs themselves and how to schedule the tasks on them in order to increase their energy efficiency. In this research, we propose a FPGA placement approach as well as a scheduling algorithm to address both aforementioned concerns. The FPGA placement mechanism employs the Heat Distribution Matrix (HDM) of servers in data center to choose host servers for FPGAs in order to reduce the effect of FPGAs on the energy consumption of CRAC as much as possible. In the next step, we propose a heuristic scheduling algorithm to assign tasks to servers and FPGAs Considering minimization of total energy consumption of IT equipment. Our joint FPGA placement and task scheduling approach can decrease the energy consumption up to 12.74% compared with original thermal obliviosu FPGA placement and task scheduling approach
  9. Keywords:
  10. Accelerators ; Data Center ; Scheduling ; Tasks ; Thermal Effect ; Energy Consumption

 Digital Object List

 Bookmark

No TOC