Flash memory based solid-state drive(SSD) has become an emerging technology and received strong interest in both academia and industry. SSDs penetrate the markets from laptops and PCs to enterprise-class server domains. SSD has its unique merit,i.e., internal parallelism; how to take advantage of this merit to improve database operations becomes an important research topic.On the other hand, a power-aware DBMS is the important requirement of database research. Thus, how to make database operationspower-aware is a non-trivial issue. In this thesis, we’ll look into these aspects for sort-merge join.The main contributions of the thesis are as follows:1. Investigate intrinsic characteristics of SSDs.This knowledge is incorporated into algorithmsconstruction for better performance on a SSD.2. With this knowledge,we proposed a novel version of parallel sort-merge join(PSMJ) whichexploits SSDs internal parallelism for high-speed data processing.3. We proposed a system-level power model, whichis referred as software-based power meter.With information feedback from hardware, we build a closed-loop power controller thatcontrols the system power by regulating the processor’s power consumption and cappingthe process power of PSMJ under power constraint.The experimental results show thatour parallel sort-merge join is faster and more energyefficiency than the traditional method and it can improve energy-efficiency by 15%. The closedloop power capping can control the power consumption in high-density power consumption environment without any expensive hardware. |