Improving Energy Efficiency and Security in Cluster Computing Systems

Abstract

Cluster computing systems are widely used in parallel and distributed computing research. Besides performance, energy cost and security should also be carefully concerned in large scale cluster computing systems to reduce budget and to avoid information leak. In this dissertation, I proposed a Time Aware Dynamic Voltage Scaling scheduling algorithm to conserve energy cost of processors in parallel computing systems and a design of an energy efficient I/O System with write buffer disks to conserve energy cost of large scale storage systems. To explain when the energy consumption could be reduced in cluster computing systems, I analyzed the CPU and I/O system performance in a security-aware storage system. Security is another issue which has not been well explored in cluster computing systems. I implemented a transparent encryption/decryption layer in a popular Message Passing Interface implementation: MPICH2. Then I quantitatively evaluate the system performance on two cluster computing systems.