Energy-Performance Tradeoffs in Information and Communication Technology

Abstract

It is of no doubt that the worldwide energy consumption and electricity costs of the Information and Communication Technology (ICT) sector has been continuously rising and the future predictions are ever increasing. Reducing this energy consumption and maintaining a cost efficient ICT sector is the focus of this dissertation.

In the first part of this dissertation, we study the energy

savings that can be achieved in data centers, most of which are underutilized, by putting servers to low power

sleeping modes. Incorporating the switching cost necessary for servers and independent cores to switch on and off,

we develop algorithms that optimally balance energy consumption and delay performance of homogeneous and

heterogeneous multicore servers.

In addition, we built a nano data center in order to experimentally

test our theoretical predictions.

We verify that our theory makes a good

prediction of the resource pooling benefits, given that the experimental results fall within 1-7% of the theoretical

ones. We also show that our approach results in higher energy savings compared to commonly adopted baseline

sleeping policies for almost all traffic loads under consideration. In particular, for traffic load 10% experimental energy savings achieved by our optimal policy reach 80%.

In the second part of this dissertation we move from intra to inter data center job scheduling.

We provide scheduling algorithms that distribute

jobs with different deadline requirements across data centers located in geographical regions characterized by different electricity

prices. Our objective is to balance completion time with transportation and electricity costs, while meeting power budget

requirements. We take into account both the space and the time dimension by allowing jobs to be scheduled across

different time slots, given that electricity prices highly fluctuate in the course of the basic scheduling period of a

day. We compare our approach with commonly adopted scheduling baseline policies and demonstrate significant

cost savings. Our results can help cloud service providers to optimally balance traffic load across data centers

so as to save substantial amounts of money spent on electricity.

In the third part of this dissertation, we extend our optimal energy conserving control techniques to the case of

DSL broadband access. By optimizing the use of heterogenous sleeping states already enabled in DSL standards, we optimally balance energy consumption and delay while demonstrating substantial energy savings.

However, DSL presents one more tradeoff related to

stability since DSL lines are usually bundled into one shielded cable. When one or more DSL lines wake up, the

crosstalk in the bundle increases, thus rendering other lines unstable. In this context, we develop techniques that

guarantee a significant improvement of the overall stability of a DSL network, resulting in a stable sleep mode

operation. We then compare our energy saving approach to that proposed by current Broadband Forum guidelines and show

that our solution can guarantee a more stable and energy efficient DSL system.

In the last part of this dissertation, we focus on the tradeoff between electricity cost and consumer's utility in a Home Area Network (HAN), as part of a Smart Grid network. We develop an optimal demand response (D/R) power scheduling algorithm,

which can run on an actual D/R controller, and with an objective of optimally balancing consumer's utility with

electricity cost. We take into account both "rich" and "poor" consumers with different money spending capabilities.

Utilizing inputs such as power consumption and operating cycles of various devices in the HAN, as well as device

priorities and real time pricing information, our algorithm schedules the operation of various devices and results

in important cost savings compared to non-D/R baseline scheduling approaches.