Abstract:
Cloud computing allows services available to customers on a pay-as-you-go
model through geographically distributed data centers (GDCs) located around the
world. Almost all Internet services are built on GDCs in order to improve
performance and reliability. As cloud computing grows in popularity, so does
energy consumption and the cost of powering servers in GDCs. Energy and cost
reduction have emerged as major challenges for GDCs.
In this thesis, energy-efficient and cost-effective resource management
framework is proposed for minimizing the operational cost and energy consumption
of servers, and ensuring the service level objective (SLO). The framework
accomplishes electricity price prediction, resource demand prediction, ensuring
SLO as well as energy-efficient and cost-effective resource allocation.
Electricity prices of GDCs in multi-regional electricity markets are
predicted. Resource demand prediction is performed by using ML techniques for
handling dynamic workload nature of the resource amount needed in data centers.
Ensuring SLO is conducted to the resource demand prediction system, to ensure the
cloud providers have enough resources to meet the resource demand. Energy
efficiency factors: allocation policies, power management techniques as well as
power models are considered to propose energy-saving resource allocation
algorithms for energy-efficient allocation of resources. Energy costs are minimized
by delivering the requests first to the data center with cheaper electricity price.
Extensive evaluations are carried out using real-world electricity price data
of GDC locations and real-world workload traces. According to the evaluation
results, the electricity price prediction model has a promising accuracy. The
resource demand prediction model predicts the accurate amount of dynamic
resource demand while meeting SLO. CloudSim is used to evaluate the performance
of the proposed resource allocation algorithms. This work contributes to reducing
energy consumption and it also offers cost-saving.
