Power management and optimization

After many years of focusing on “faster” computers, people have started taking notice of the fact that the race for “speed” has had the unfortunate side effect of increasing the total power consumed, thereby increasing the total cost of ownership of these machines. The heat produced has required expensive cooling facilities. As a result, it is difficult to ignore the growing trend of “Green Computing,” which is defined by San Murugesan as “the study and practice of designing, manufacturing, using, and disposing of computers, servers, and associated subsystems – such as monitors, printers, storage devices, and networking and communication systems – efficiently and effectively with minimal or no impact on the environment”. There have been different approaches to green computing, some of which include data center power management, operating system support, power supply, storage hardware, video card and display hardware, resource allocation, virtualization, terminal servers and algorithmic efficiency. In this thesis, we particularly study the relation between algorithmic efficiency and power consumption, obtaining performance models in the process. The algorithms studied primarily include basic linear algebra routines, such as matrix and vector multiplications and iterative solvers. Our studies show that it if the source code is optimized and tuned to the particular hardware used, there is a possibility of reducing the total power consumed at only slight costs to the computation time. The data sets utilized in this thesis are not significantly large and consequently, the power savings are not large either. However, as these optimizations can be scaled to larger data sets, it presents a positive outlook for power savings in much larger research environments