This thesis describes an approach to add functionality and improved performance to the Hadoop infrastructure for cloud computing. In particular, we have added code to the Hadoop source files, to allow unix scripts to run on the task nodes of the cloud, from within the mapper phase of Hadoop execution. Our results show that the new approach is easier to program than other alternatives, more easy to understand for experienced UNIX programmers, more powerful in terms of the kinds of computations that are possible, and as fast or faster to compute than would be the alternatives.

More and more clouds with diversified properties have been built. Many of them span multiple geographical locations over the globe, imposing time-varying costs on and offering different service proximities to users. Clouds could be private or public, requiring different levels of administration effort and providing different levels of freedom to control. Hybrid clouds, which blend together multiple public and private clouds, possess properties of both types. Based on these diversified properties, multiple clouds have the potential to provide services with higher scalability, lower operational cost and better QoS. To exploit this potential, I examine the means of deploying services on multiple clouds, that can maximize utility in dynamic environments. Firstly, I consider the migration of an important representative application, content distribution services, to geo-distributed hybrid clouds. I model the problem of joint content data migration and request dispatching as a unified optimization framework, and then design a dynamic control algorithm to solve it. The algorithm bounds the response times within the preset QoS target, and guarantees that the overall cost is within a small constant gap from the optimum that can be achieved by a T-slot look ahead mechanism with known future information. Secondly, I study the problem of efficient scheduling for disparate MapReduce workloads on hybrid clouds. I build a fine-grained and tractable model to characterize the scheduling of heterogeneous MapReduce workloads. An online algorithm is proposed for joint task admission control for the private cloud, task outsourcing to the public cloud, and VM allocation to execute the admitted tasks on the private cloud, such that the time-averaged task outsourcing cost is minimized over the long run. The online algorithm features preemptive scheduling of the tasks, where a task executed partially on the on-premise infrastructure can be paused and scheduled to run later. It also achieves such desirable properties as meeting a pre-set task admission ratio and bounding the worst-case task completion time. Thirdly, I consider a cloud computing resource market where a broker is employed that pools the spare resources of multiple private clouds and leases them to serve external users' jobs. I model the interaction between the broker and the private clouds as a two-stage Stackelberg game. As the leader in the game, the broker decides on the pricing for renting VMs from each private cloud. As a follower, each private cloud responds with the number of VMs that it is willing to lease. Combining all this with Lyapunov optimization theory, I design online algorithms for the broker to set the prices and schedule the jobs on the private clouds, and for each private cloud to decide the numbers of VMs to lease. The broker achieves a time-averaged profit that is close to the offline optimum with complete information on future job arrivals and resource availability, while each private cloud earns the best that it can. Through theoretical analysis and empirical study, I rigorously examine the cost or profit optimality, and QoS guarantee of my design, and show that they can indeed outperform existing solutions. / published_or_final_version / Computer Science / Doctoral / Doctor of Philosophy

No description available.

No description available.

Throughout the recent years cloud computing has gained large popularity in the information technology domain. Despite its popularity, there are many organizations that are lacking broader understanding of implementing and utilizing cloud computing for business and operating purpose due to the existing vagueness regarding its cost and security effect associated. It is argued that the main attractiveness of cloud computing for organizations is its cost effectiveness, whilst the major concern relates to the risks for security. Accordingly, more effort has been made in exploring these issues of cloud computing impact. However, little effort has been focused at critically examining the cost risks and security benefits which cloud computing bring to organizations. By using a qualitative method this research examines in detail the essential benefits and risks of cloud computing utilization for organizations in terms of cost and security. Unlike prior studies, it also explores the cost risks and security benefits and shows that they should be taken into consideration by organizations. The findings are based on empirical data collected via interviews with IT professionals. The main cost risk identified is the lack of accurate and sophisticated cost models on the current cloud market. Among the identified security benefits are increased data safety, faster data recovery and transfer, centralization, and improved security software mechanisms and maintenance. Moreover, this research shows several major implications that organizations should keep in mind while utilizing cloud computing and provides some suggestions on how to avoid the cost and security risks identified. At present, reduction of the operational and administrative costs is seen by organizations as the most essential cost benefit. The results show that cloud computing is better for small- and medium-sized organizations and that the hybrid cloud is the most appropriate model for them. Furthermore, the cost and security risks of cloud computing cannot be avoided without resolution of the problem with the lack of accurate cost models, international regulatory frameworks and interoperable security standards on supranational levels.

With cloud computing emerging in recent years, more and more interest has been sparked from a variety of institutions, organizations and individual users, as they intend to take advantage of web applications to share a huge amount of public and private data and information in a more affordable way and using a reliable IT architecture. In the area of healthcare, medical and health information systems based on cloud computing are desired, in order to realize the sharing of medical data and health information, coordination of clinical service, along with effective and cost-contained clinical information system infrastructure via the implementation of a distributed and highly-integrated platform. The objective of this study is to discuss the challenges of adopting cloud computing for collaborative health research information management and provide recommendations to deal with corresponding challenges. More specially, the study will propose a cloud computing based platform according to recommendations. The platform can be used to bring together health informatics researchers from the different geographical locations to share medical data for research purposes, for instance, data mining used for improving liver cancer early detection and treatment. Finding from a literature review will be discussed to highlight challenges of applying cloud computing in a wide range of areas, and recommendations will be paired with each challenge. A proof of concept prototype research methodology will be employed to illustrate the proposed cross national cloud computing model for geographically distributed health data mining applied to a health informatics research. / Graduate / 0573

No description available.

Proefschrift--Utrecht. / Summary in English, French and German. "Litteratuur": p. 92.

Proefschrift--Utrecht. / Summary in English, French and German. "Litteratuur": p. 92.

Thesis--Wisconsin. / Includes bibliographical references (leaves 110-114).