Power Aware Meta Scheduler for Adaptive VM Provisioning in IaaS Cloud

Introduction

With the advent of cloud computing, large scale data centers are becoming common in the computing industry. However, these data centers equipped with high performance infrastructures consume huge power causing global warming by emitting CO₂ footprint, giving a serious environmental threat to today’s world. One of the major causes for energy inefficiency in the data center is the idle power wasted when servers run at low average utilization. Even at 10% of CPU utilization, the power consumed is over 50% of the peak power (Neugebaur & McAuley, 2001; Ragavendra et al., 2008). This results in more power consumption per workload during off-peak load. Pinheiro and Rajamony quoted that 22% of energy consumption of a single server is needed to cool it. A study on data center issues also shows that energy consumption of data centers worldwide doubled between 2000 and 2006 (Pinheiro et al., 2001; Elnozahy, 2003). Incremental US demand for data center energy between 2008 and 2010 is the equivalent of 10 nuclear power plants (Kaplan et al., 2008). To handle this issue, data centers perform consolidation of various workloads onto a set of common servers with the help of live state migration facility which is enabled through virtualization technology (James & Ravi, 2005).

In a cloud environment, the power and energy management strategies need to consider the characteristics of servers and incoming workloads. Modern microprocessor technology supports the processing elements (PE) such as core, chip and host to be set in different sleep states depending on the demand (Lee et al., 2007). The sleep states are also referred to as power saving states. A PE conserves different amount of power at different sleep states. The power drawn during wake up time is comparatively insignificant. The shallow sleep states realize lower power conservation with lower wake up latency and the deep sleep states realize higher power conservation with higher wake up latency. IBM’s Power family machines support nap and sleep modes (Sinharoy et al., 2005; Kim et al., 2011; Cardosa et al., 2009). The nap is a low-power state designed for short processor idle periods (Malcolm et al., 2010). It provides modest power reduction over a software idle loop, and the wake-up latency of the nap state is less than 5μs. Instruction execution begins immediately upon wakeup. The second idle mode is sleep state. It is a lower-power, higher latency standby state intended for processing cores that will be unused for an extended period of time.