A Dynamic Power Management Schema for Multi-Tier Data Centers

Abstract

An issue of great concern as it relates to global warming is power consumption and efficient use of computers especially in large data centers. Data centers have an important role in IT infrastructures because of their huge power consumption. This thesis explores the sleep state of data centers' servers under specific conditions such as setup time and identifies optimal number of servers. Moreover, their potential to greatly increase energy efficiency in data centers. We use a dynamic power management policy based on a mathematical model. Our new methodology is based on the optimal number of servers required in each tier while increasing servers' setup time after sleep mode to reduce the power consumption. The Reactive approach is used to prove the validity of the results and energy efficiency by calculating the average power consumption of each server under specific sleep mode and setup time. We introduce a new methodology that uses average power consumption to calculate the Normalized-Performance-Per-Watt in order to evaluate the power efficiency. Our results indicate that the proposed schema is beneficial for data centers with high setup time.