Thermal-aware scheduling in green data centers / Muhammad Tayyab Chaudhry

• Main Author: Chaudhry, Muhammad Tayyab
• Format: Thesis
• Published: 2015
• Subjects: QA75 Electronic computers. Computer science; QA76 Computer software
• Online Access: http://studentsrepo.um.edu.my/5814/1/WHA110024_PhD_Thesis_Thermal%2Daware_Scheduling_in_Green_Data_center.pdf; http://studentsrepo.um.edu.my/5814/

Data centers can go green by saving electricity in two major areas: computing and cooling. Servers in data centers require a constant supply of cold air from on-site cooling mechanism for reliability. Increased computational load makes servers to dissipate more power as heat and eventually amplifies the cooling load. In thermal-aware scheduling, computations are scheduled with the objective of reducing data center wide thermal gradient, hotspots and cooling magnitude. Complemented by heat modeling, thermal-benchmarking, thermal-aware server arrangement; and thermal-aware monitoring and profiling, this scheduling is energy efficient and economical. This research work proposes multiple techniques for thermal-benchmarking of data center servers such as: Thermal-benchmarking for Standalone Servers (TBSS), Thermal-benchmarking for Server Comparison (TBSC), Multi-intensity TBSS (MiTBSS) and Thermal-benchmarking for Virtualized Clusters (TBVC). These techniques are useful for thermal evaluation of servers, emulating various types of workloads and creating the thermal profiles. A thermal-aware server relocation algorithm (ThSRA) for thermal-stress free arrangement of servers is also proposed. The experimental results show that the peak outlet temperatures of the servers can be brought closer to average outlet temperature by over 5 times through ThSRA. This also brings the lowering of average peak outlet temperature by 3.5% and minimizing the thermal-stress. Thermal profiles are used for outlet temperature prediction modeling of the servers. These models include the worst case prediction model (WCPM), optimistic prediction model (OPM) and enhanced optimistic prediction model (EOPM). The best prediction model can predict the outlet temperature of the servers with an average error of up to 0.3 degree Celsius. WCPM is applied for offline hotspot-resistant virtual machine deployment algorithm (HVMDA) and hotspot-aware server arrangement algorithm (HSLERA). The combination of HVMDA and HSLERA leads to increase in server utilization by up to 50% and lowering the peak outlet temperature by up to 3% on average. The WCPM and OPM are used for the implementation of online thermal-aware VM scheduling. These schedulers have comparatively lower thermal-gradient across all servers, lower outlet temperatures across all servers, effective use of computing capacity and the power consumption. The proposed proactive schedulers comparatively show up to 11% in total energy savings. All these thermal-aware techniques are helpful in the establishment of green data centers.