Economic and environmental sustainability for cloud computing systems : strategies and algorithms

With the enormous growth of cloud computing technology the data centers, which serve as the primary physical resources to provide various services, are also growing in size and complexity at a tremendous pace. This thesis aims to contribute towards the two important issues faced by cloud service operators; namely, i) economic sustainability and ii) environmental sustainability. Firstly, the scenario of computing the optimal service prices for a cloud service provider operating in a competitive cloud service market, with multiple cloud service providers, is considered. In this paradigm, all the cloud service providers operating in such markets compete among themselves to attract customers. The market consists of price sensitive, rational customers. As a result, eventually all the customers will move to that cloud service provider, which offers better service quality at a lower cost. Hence, each cloud service provider needs to strategically determine the optimal set of actions (including the service quality offered and the price charged for the same). These actions have to consider the set of actions chosen by the other competitors, operating in the same market. This will ensure a healthy competition among the cloud service providers and will eliminate the possibility of formation of a monopolistic market. The contribution of this thesis is that, through a bi-level optimization strategy, the environmental sustainability issue (through renewable energy integration) is connected to the long term economic sustainability issue, which deals with the revenue earned and the market share in terms of number of customer requests served. Pricing decisions are taken in the slower timescale and energy efficient job scheduling is performed in the faster timescale. To enable the above mentioned bi-level optimization strategy a modeling framework is provided to determine the appropriate pricing strategy based on the preferences strategically chosen by each of the cloud service providers. Through this framework the environmental sustainability issue is addressed by appropriately pricing the renewable energy. All the proposed algorithms are evaluated with real life data traces for electricity price, workload, renewable generation and all the other model parameters. Then, the case of a single data center is considered where different applications are deployed in a tiered architecture, which is usually the case of real systems. Two job scheduling algorithms are developed and a rigorous investigation is carried out to understand their performance. Under this framework, thereafter a novel incentivization strategy to achieve more aggressive renewable energy integration and electricity cost reduction is designed. The proposed algorithms are evaluated with real life data traces for electricity price, workload, renewable generation and all the other model parameters. Then the case of a cloud computing system with multiple data centers is considered. A rigorous investigation is carried out on the load distribution strategy for a cloud service provider owning multiple geographically distributed data centers, which is the case for a typical mid to large cloud service provider. In this case two algorithms for optimal load distribution are presented and compared. In order to hedge the risk associated with the volatility of average cost of electricity a simple derivate, namely forward contract, is designed and implemented. In order to address the economic sustainability issue a few relevant topics in economics (such as price sensitivity, market equilibria, competition etc.) are leveraged. To address the environmental sustainability issue, a promising technique prevalent in the smart grid arena namely, demand response, is leveraged. Since the cloud computing technology is still in a nascent stage, the strategies and algorithms proposed in this thesis will indeed make some meaningful contributions in shaping the future of cloud computing.DOCTOR OF PHILOSOPHY (EEE