This thesis describes a method of reducing communication overheads within the MapReduce infrastructure of a cloud computing environment. MapReduce is an framework for parallelizing the processing on massive data systems stored across a distributed computer network. One of the benefits of MapReduce is that the computation is usually performed on a computer (node) that holds the data file. Not only does this approach achieve parallelism, but it also benefits from a characteristic common to many applications: that the answer derived from a computation is often smaller than the size of the input file. Our new method benefits also from this feature. We delay the transmission of individual answers out a given node, so as to allow these answers to be combined locally, first. This combination has two advantages. First, it allows for a further reduction in the amount of data to ultimately transmit. And second, it allows for additional computation across files (such as a merge-sort). There is a limit to the benefit of delaying transmission, however, because the reducer stage of MapReduce cannot begin its work until the nodes transmit their answers. We therefore consider a mechanism to allow the user to adjust the amount of delay before data transmission out of each node.

Cloud computing is a new concept that incorporates many existing technologies, such as virtualization. Virtualization is important for the establishment of cloud computing. With virtualization, cloud computing can virtualize the hardware resources into a huge resource pool for users to utilize. This thesis begins with an introduction to how a widely used service model classifies cloud computing into three layers. From the bottom up, they are IaaS, PaaS, and SaaS. Some service provides are taken as examples for each service model, such as Amazon Beanstalk and Google App Engine for PaaS; Amazon CloudFormation and Microsoft mCloud for IaaS. Next, we turn our discussion to the hypervisors and the technologies for virtualizing hardware resources, such as CPUs, memory, and devices. Then, storage and network virtualization techniques are discussed. Finally, the conclusions and the future directions of virtualization are drawn.

Cloud computing becomes popular nowadays. It allows applications to use the enormous resources in the clouds. With the combination of mobile computing, mobile cloud computing is evolved. With the use of clouds, mobile applications can offload tasks to clouds in client-server model. For cloud computing, migration is an important function for supporting elasticity. Lightweight and portable task migration support allows better resource utilization and data access locality, which are essentials for the success of cloud computing. Various migration techniques are available, such as process migration, thread migration, and virtual machine live migration. However, for these existing migration techniques, migrations are too coarse-grained and costly, and this offsets the benefits from migration. Besides, the migration path is monotonic, and mobile and clouds resources cannot be utilized. In this study, we propose a new computation migration technique called stack-on-demand (SOD). This technique is based on the stack structure of tasks. Computation migration is carried out by exporting parts of the execution state to achieve lightweight and flexible migration. Compared to traditional task migration techniques, SOD allows lightweight computation migration. It allows dynamic execution flows in a multi-domain workflow style. With its lightweight feature, tasks of a large process can be migrated from clouds to small-capacity devices, such as iPhone, in order to use the unique resources, such as photos, found in the devices. In order to support its lightweight feature, various techniques have been introduced. To allow efficient access to remote objects in task migration, we propose an object faulting technique for efficient detection of remote objects. This technique avoids the checking of object status. To allow portable, lightweight application-level migration, asynchronous migration technique and twin method hierarchy instrumentation technique are proposed. This allows lightweight task migration from mobile device to cloud nodes, and vice versa. We implement the SOD concept as a middleware in a mobile cloud environment to allow transparent execution migration of Java programs. It has shown that SOD migration cost is pretty low, comparing to several existing migration mechanisms. We also conduct experiments with mobile devices to demonstrate the elasticity of SOD, in which server-side heavyweight processes can run adaptively on mobile devices to use the unique resources in the devices. On the other hand, mobile devices can seamlessly offload tasks to the cloud nodes to use the cloud resources. In addition, the system has incorporated a restorable communication layer, and this allows parallel programs to communicate properly with SOD migration. / published_or_final_version / Computer Science / Doctoral / Doctor of Philosophy

Cloud computing, which is among the trendiest computing paradigms in recent years, is believed to be most suitable for supporting network-centric applications by providing elastic amounts of bandwidth for accessing a wide range of resources on the y. In particular, geo-distributed cloud systems are widely in construction nowadays. They span multiple data centers at different geographical locations, thus offering many advantages to large-scale multimedia applications because of the abundance of on-demand storage/bandwidth capacities and their geographical proximity to different groups of users. In this thesis, we investigate the common fundamental challenges in how to efficiently leverage the power of cloud resources to facilitate multimedia content delivery in various modern real world applications, from different perspectives. First, from the perspective of application providers, we propose tractable procedures for both model analysis and system designs of supporting representative large scale multimedia applications in a cloud system, i.e., VoD streaming applications and social media applications, respectively. We further verify the effectiveness of these algorithms and the feasibility of their deployment under dynamic realistic settings in real-life cloud systems. Second, from the perspective of end users, we target our focus at mobile users. The rapidly increasing power of personal mobile devices, dwarfing even high-end devices, is providing much richer contents and social interactions to users on the move, and many more challenging applications are on the horizon. We explore the tough challenges of how to effectively exploit cloud resources to facilitate mobile services by introducing two cloud-assisted mobile systems (i.e., CloudMoV and vSky-Conf), and explain in details their design philosophies and implementation. Finally, from the perspective of the cloud providers, we realize existing data center networks lack the flexibility to support many core services, given our hands-on experiences from working with public cloud systems. One of the specific problem is, “bulk data transfers across geo-distributed datacenters". After formulating a novel and well-formed optimization model for treating the data migration problem, we design and implement a Delay Tolerant Migration (DTM) system based on the Beacon platform and standard OpenFlow APIs. The system realizes a reliable Datacenter to Datacenter (D2D) network by applying the software defined networking (SDN) paradigm. Real-world experiments under realistic network traffic demonstrate the efficiency of the design. / published_or_final_version / Computer Science / Doctoral / Doctor of Philosophy

The cloud, in the past few years, has become the preferred platform for hosting web applications. Many of these web applications store their data in a distributed cloud storage system, which greatly simplifies application development and provides increased availability and reliability. However, with increasing user demand for web applications, these cloud storage systems often become the performance bottleneck. To address the cloud's performance demands, many storage system features, such as strong consistency and transactional support, are often omitted in favour of performance. Nonetheless, transactions remain necessary to ensure data integrity and application correctness. In this thesis, we introduce CrossStitch, which is an efficient transaction processing framework for distributed key-value storage systems. CrossStitch supports general transactions, where transactions include both computation and key accesses. It is specifically optimized for short-lived transactions that are typical of cloud-deployed web applications. In CrossStitch, a transaction is partitioned into a series of components that form a transaction chain. These components are executed and the transaction is propagated along the storage servers instead of being executed on the application server. This chained structure, in which servers only communicate with their immediate neighbours, enables CrossStitch to implement a pipelined version of two-phase commit to ensure transactional atomicity. CrossStitch is able to eliminate a significant amount of setup overhead using this structure by executing the transaction and the atomic commit protocol concurrently. Therefore, CrossStitch provides low latency and efficient transactional support for cloud storage systems. Our evaluation demonstrates that CrossStitch is a scalable and efficient transaction processing framework for web transactions.

Ce travail doctoral s’intéresse aux impacts sécuritaires de l’usage du cloud computing dans les petites et moyennes entreprises (PME). Il repose sur une méthodologie de recherche mixte, à la fois qualitative et quantitative, en adoptant une posture épistémologique orientée réalisme critique (Mingers, 2002,2004). Son objectif est de mettre en lumière les éléments d’une théorie d’explication et de prédiction (Gregor, 2006) de la portée sécuritaire de l’adoption du cloud computing dans les petites et moyennes entreprises. Les résultats de notre recherche se matérialisent au travers la présentation et la publication de quatre travaux académiques : Un premier article publié (un modèle d’acheminement des données depuis les PME vers ses fournisseurs cloud potentiels), deux articles acceptés pour publication (un modèle du transfert des risques SI dans un écosystème cloud computing, une mise en lumière de la délégation du rôle du DSI) et un dernier article en cours d’évaluation (la caractérisation de la réversibilité d’un service cloud). La thèse contribue à une meilleure compréhension des conséquences sécuritaires inter- et intra-organisationnelles, entre autres à travers la mobilisation des théories de la firme et la modélisation mathématique. Elle propose, en outre, un nouveau coefficient d’estimation de la concordance pour des hypothèses indépendantes, / This doctoral work focuses on security impacts of the cloud computing use in SMEs. It is elaborated through a mixed research methodology by adopting a critical realism position (Mingers, 2002, 2004). Its objective is to implement a theory of explanation and prediction (Gregor, 2006) applied to the safe scope of the cloud computing adoption in small and medium enterprises. The results of our research are materialized through the publication of an article (a model of data transmission from SMEs to its potential cloud suppliers) two articles accepted for publication (a model of the transfer of risks SI in a cloud computing ecosystem, highlighting the delegation of the role of the CIO) and an article under evaluation (the characterization of the reversibility of a cloud service). The thesis contributes to a better understanding of the inter- and intra-organizational security consequences, among others through the mobilization of firm theories and the mathematical modeling. It also proposes a new coefficient of agreement estimation for independent assumptions.

Managing information security in the cloud is a challenge. Traditional checklist approaches to standards compliance might well provide compliance, but may not provide adequate security assurance. The complexity of cloud relationships must be acknowledged and explicitly managed by recognising the implications of the self-interest of each party involved. We develop a conceptual modelling framework for cloud security assurance that can be used as a starting point for achieving effective continuous security assurance, together with a high level of compliance.

The aim of my diploma thesis is to introduce cloud computing as an alternative to traditional internal information technology and its benefits for a company. Diploma thesis focuses on three main goals. The first one concerns advantages and disadvantages of cloud computing in comparison to internal information technology. The second one is identification of possible processes and services available for migration into cloud computing. The third goal of the diploma thesis is development of investment analysis which compares not only initial costs on internal information technology and cloud computing, but also costs of both variants within three years. The main contribution of the diploma thesis is to define whether the cloud computing is economically beneficial for the company or not. The argument for categorizing cloud computing is in the reason that the company does not have to use all services within the cloud computing but only these, which the company considers as the most beneficial from cost and operation point of view. Another contribution of the diploma thesis is deployment of data, services and processes into a chosen cloud computing platform. Investment analysis allows through cost comparison of both options understand whether it is more beneficial to choose cloud computing or internal Information Technology platform. During this decision making the company also considers its business character and the fact whether the company operates locally or globally.

The current infrastructure as a service (IaaS) cloud systems, allow users to load their own virtual machines. However, most of these systems do not provide users with an automatic mechanism to load a network topology of virtual machines. In order to specify and implement the network topology, we use software switches and routers as network elements. Before running a group of virtual machines, the user needs to set up the system once to specify a network topology of virtual machines. Then, given the user’s request for running a specific topology, our system loads the appropriate virtual machines (VMs) and also runs separated VMs as software switches and routers. Furthermore, we have developed a manager that handles physical hardware failure situations. This system has been designed in order to allow users to use the system without knowing all the internal technical details.

With the emergence of cloud computing, many applications are migrated onto clouds. Video-on-demand (VoD) can be implemented on the cloud platform with geo-distributed cloud servers to serve worldwide users. New videos are distributed to these geo-distributed cloud servers. This distribution should be properly scheduled based on the videos' sizes, videos' popularities and the available network bandwidth, so that the mean completion time is minimized. We formulate this problemas a preemptive scheduling problem, prove that it is NP-hard, and design a heuristic scheduling algorithm to solve it. This algorithm iteratively determines: 1) themost preferred file to be received by the most preferred destination server by pairwise analysis and PageRank, and 2) the most preferred source servers which can transmit this file to the most preferred destination server with appropriate data rates. The simulation results demonstrate that the proposed scheduling algorithm gives much smaller mean completion time than four scheduling algorithms.