In my thesis project, I investigate buffering schemes (Separate Queue, Class Queue, and Common Queue) in a generalized cluster model of machines. A ranking method for physical computing elements is based on the SPECpower_ssj2008 benchmark of Standard Power Evaluation Corporation. The impact of buffering schemes is evaluated on system performance and energy conservation, when either SQEE (Shortest Queue with Energy Efficiency priority) or SQHP (Shortest Queue with High Performance priority) scheduling policy is applied.
The experimental results showed that the proposed buffering schemes (i.e. Class Queue and Common Queue) outperform the traditional one (Separate Queue) without impact on energy consumption. Additionally, motivated from existing drawbacks of applied scheduling policies, dynamic scheduling algorithms are proposed to balance between performance capacity and energy efficiency issue. The new algorithms are examined in the common queue cluster model, which is demonstrated as the best performance configuration.
We also examined the DVFS (Dynamic Voltage and Frequency Scaling) technique for purpose of saving energy in every studied model. The obtained results present the inconsistent impacts in different cases of study. Therefore, it is very important to note that DVFS should be critically examined if one would like to apply it in a practical scheduling model.