Authors : G. Ravikumar and N. Nagarajan
Abstract: Power consumption of large servers and disks has become a popular research topic as this issue is important from both technical and environmental perspectives. The performance of the disk systems are greatly affected by extreme power consumption. A majority of the research in disk power management has concentrated on the behavior of the disk during periods of idleness. The main focus is on when the disk should be put to idleness to reduce power consumption without affecting the performance. Due to the increasing requirements of current and forthcoming data-intensive computer applications, there has been a chief alteration in the disk subsystem which now comprises of more disks with higher storage capacities and higher rotational speeds. Thus, disk power management has become a vital issue as it consumes very high power. This study proposes and evaluates an efficient compiler-directed disk power management technique which utilizes disk access schemes for reducing energy consumption. This study uses a novel approach called LMFNN in which the Fuzzy Neural Network is trained using Modified Levenberg Marquardt Learning algorithm. The proposed scheme analyzes the various disk access techniques and selects the suitable algorithm which would provide better overall performance of the disks. The experimental evaluation using a diverse set of workloads indicates that the proposed LMFNN approach provides better power consumption than the conventional approaches.
G. Ravikumar and N. Nagarajan, 2012. A Fuzzy Neural Network Based Compiler Directed Power Optimization for Disk Based Systems with Modified LM Training. International Journal of Soft Computing, 7: 283-293.