Energy-aware Memory Management through Database Buffer Control

Energy consumption in modern data center trends to be increasing, which gives pressure to limit power consumption in servers. There is an opposite side trend that needs more memory resources to load increasing application data. Once a system is designed with capacity for peak time usage, it inefficiently consumes more energy in an idle state. So in this paper we propose a dynamically adaptie applicationdriven memory management scheme. Our example is on a database application with an online transactional workload. We design and implement a simple heuristic scheme, which determines the state of either expanding or shrinking the size of memory. Our experimental results show that significant amount of energy saving (4–8%) is achieved without any visible performance penalty. Although we acknowledge that the overall amount of energy reduction varies to the amount of memory in a system, we believe that this is equally achievable over high-end server machines that are similar to our test machine.