ارزیابی عملکرد و مصرف انرژی یک سیستم شبکه ناهمگن در زمان واقعی با استفاده از DVS و DPM

Energy consumption of large scale systems has been severely studied due to economic and ecological reasons. This paper studies energy gains that come from the application of two popular energy saving techniques, Dynamic Voltage Scaling (DVS) and Dynamic Power Management (DPM), in a real-time 2-level heterogeneous grid system. While these techniques generally work in a competitive way, we show that under certain circumstances they can work together and achieve greater savings when they are both applied at the processor level. A simulation model is used to evaluate the performance of the system. Experimental results show encouraging energy savings up to 46% and minimum performance degradation when both energy saving techniques are applied.

Energy consumption of computer systems has recently gained significant attention due to the following reasons: • A system that consumes less energy, generates less heat and therefore has a longer lifetime. According to a formula based on the Arrhenius Law, component life expectancy decreases 50% for every 10 °C temperature increase. Thus, reducing a component’s operating temperature by the same amount (consuming less energy), doubles its life expectancy [1]. • In battery operated systems, a crucial objective is to work under the minimum power consumption and provide a desired level of performance at the same time. Portable devices need to preserve energy in order to increase autonomy. • The heat in large scale computers requires expensive cooling technologies. In order for large companies to reduce cooling cost they consider building large server farms in cold places. Facebook is to build a multi-million server farm on the edge of the Arctic Circle in Northern Sweden. The center will be managing all of Facebook’s traffic in Europe, serving more than 180 million users. • Electricity cost of the computing hardware is very high [2]. Estimations show that computing center servers currently consume about 0.5% of the total electricity consumption in the world, and the percentage will be quadruple by 2020 if the current demand keeps rising [3]. • Environmental reasons. The global information and communications technology (ICT) industry accounts for approximately 2% of global carbon dioxide (CO2) emissions, according to an estimate by Gartner Inc. [4].

In this paper we studied the performance of scheduling policies and the efficiency of adaptive power-saving mechanisms when applied to a real time 2-level grid system that consists of four heterogeneous clusters. Our objective was to schedule tasks and apply DVS and DPM to the processor level, in such a way that as much as possible deadlines are satisfied and energy consumption is reduced. We managed to achieve 15–46% energy reduction while deadline miss ratio was up to 0.5% for high priority tasks. When the system’s load was low, we achieved 46% energy reduction while for high arrival rates (high system load) energy gains achieved were 15%. We conclude that processors’ energy consumption can be further reduced by applying DPM when the load of the system is relatively low. We intend to extend our research by adding components and peripheral devices and study not only processors’ energy consumption, but system’s overall energy consumption. Heterogeneity of the system can be further explored if mobile devices with low performance and low power consumption are added to the system.