Eficiência energética e qualidade de serviço em centros de dados

Abstract

Data centers are among the largest consumers of electric energy in the information technology area. The energy consumption of a single data center can be larger than the consumption of a city with 100 thousands inhabitants. Data centers are not only famous for consuming large quantities of energy but also for being inefficient. Recent studies show that, on average, more than 50% of the energy consumed by a data center are wasted because of inefficient management of its resources. The energy consumption of a data center is dominated by three dimensions: servers, cooling system, and network infrastructure. The current proposals for energy management, generally, try to optimize one or two of these dimensions without considering the others. Moreover, they do not consider how these energy reduction optimizations impact performance parameters of the data center. This work first proposes an architecture for management and simulation of data centers for energy efficiency studies. Based on this architecture, this work develops the simulator SimDC3D that simulates realistically the subsystems that consume the most energy in a data center and how algorithms for scheduling and relocation of virtual machines, controlling the cooling system, and optimizing the network topology interfere with the performance parameters of the data center. This work also proposes eight new algorithms for scheduling and relocation of virtual machines, one algorithm for controlling the cooling system, and one algorithm for optimizing the network topology of the data center. The scheduling and relocation algorithms are highly parameterized so that a data-center operator can optimize one of the dimensions more aggressively than the others, i.e., he can assign different weights to the subsystems. Moreover, this work does an extensive evaluation, with hundreds of simulations, of the proposed algorithms under different workloads and topologies and compares them with the existing proposals. The results show the importance of setting the appropriate parameters for the optimization algorithms so that they can not only save energy but also avoid violations of service-level agreements settled between data center operators and their customers.