Abstract

The unexpected growth, flexibility and dynamism of information technology (IT) over the last
decade has radically altered the civilization lifestyle and this boom continues as yet. Many nations
have been competing to be forefront of this technological revolution, quite embracing the opportunities
created by the advancements in this field in order to boost economy growth and to increase the
accomplishments of everyday’s life. Cloud computing is one of the most promising achievement of
these advancements. However, it faces many challenges and barriers like any new industry. Managing
and maximizing such a very complex system business revenue is of paramount importance.
The wealth of the cloud protfolio comes from the proceeds of three main services: Infrastructure as
a service (IaaS), Software as a service (SaaS), and Platform as a service (PaaS).

The Infrastructure as a Service (IaaS) cloud industry that relies on leasing virtual machines (VMs)
has a significant portion of business values. Therefore many enterprises show frantic effort to capture
the largest portion through the introducing of many different pricing models to satisfy not
merely customers’ demands but essentially providers’ requirements. Indeed, one of the most challenging
requirements is finding the dynamic equilibrium between two conflicting phenomena: underutilization
and surging congestion. Spot instance has been presented as an elegant solution to
overcome these situations aiming to gain more profits. However, previous studies on recent spot
pricing schemes reveal an artificial pricing policy that does not comply with the dynamic nature of
these phenomena.

In this thesis, we investigate dynamic pricing of stagnant resources so as to maximize cloud revenue.
To achieve this task, we reveal the necessities and objectives that underlie the importance of
adopting cloud providers to dynamic price model, analyze adopted dynamic pricing strategy for real
cloud enterprises and create dynamic pricing model which could be a strategic pricing model for
IaaS cloud providers to increase the marginal profit and also to overcome technical barriers simultaneously.

First, we formulate the maximum expected reward under discrete finite-horizon Markovian decisions
and characterize model properties under optimum controlling conditions. The initial approach
manages one class but multiple fares of virtual machines. For this purpose, the proposed approach
leverages Markov decision processes, a number of properties under optimum controlling conditions
that characterize a model’s behaviour, and approximate stochastic dynamic programming using linear
programming to create a practical model.

Second, our seminal work directs us to explore the most sensitive factors that drive price dynamism
and to mitigate the high dimensionality of such a large-scale problem through conducting column
generation. More specifically we employ a decomposition approach.

Third, we observe that most previous work tackled one class of virtual machines merely. Therefore,
we extend our study to cover multiple classes of virtual machines. Intuitively, dynamic price
of multiple classes model is much more efficient from one side but practically is more challenging
from another side. Consequently, our approach of dynamic pricing can scale up or down the price
efficiently and effectively according to stagnant resources and load threshold aims to maximize the
IaaS cloud revenue.