Memory power-aware virtual machine scheduling

Abstract

Recent technology advance in hardware and software resurrects the virtualization era. Especially, server consolidation using virtualization is widely employed in large-scale computing facilities and data centers. Increasing energy consumption in server consolidation environments leads to high maintenance costs as well as environmental problems. Main memory, no less than processor, is a major energy consumer in this environment. DRAM is used as main memory for most of computer systems currently available. The DRAM hardware already provides power-saving functionality. Using this feature, several previous researches successfully save energy consumed in the DRAM. This paper proposes a technique for reducing memory energy consumption using virtual machine scheduling in multi-core systems. Previously, in a single processor system, a virtual machine runs at a specific time and the memory used by the virtual machine is the only source of memory energy consumption. The appearance and popularity of multi-core systems, however, introduce additional problems in reducing memory energy consumption; several virtual machines simultaneously run on each processor core and the memory energy consumption is the aggregated memory energy used by all these virtual machines. Thus, it is a challenging issue to schedule virtual machines that access as many same memory nodes as possible in order to save more memory energy. We devise several heuristic scheduling algorithms such as BVF, BCSF, BNF, and COMB. A memory power simulator called MPSim is designed and implemented. In the simulation results, we observe that our memory power management effectively saves memory power. We also find out that more memory energy savings are gained as 1) system size increases, 2) the memory nodes used by a virtual machine decreases, and 3) the more number of virtual machines are consolidated. Among the devised algorithms, BCSF shows the best energy saving performance in the most simulation configurat...