Optimized Strategies Of Cloud-Assisted File Sharing&Releasing Systems

With the rapid development of cloud-computing techenologies, an increasing number of Internet applications emerge with cloud platform. As popular services with a large amount of users, P2P (peer-to-peer) file distribution systems are also evoluting toward cloud, and then turning into to the cloud-assisted file sharing&releasing systems.The hybrid file sharing&releasing system which combines P2P mode and cloud serving mode, reflects many novel features such as scalable bandwidth reservation from cloud, providing download appointment service to users, providing both peer-assisted acceleration and cloud-assisted acceleration to download processes, and alleviating impacts of flash crowd during the start-up phase of releasing process of popular content by short-term renting cloud resources, and so on. Driven by the novel serving mode and new characteristics, many issues must be tackled, specifically about performance improvement and cost saving of file sharing&releasing systems,Firstly, does the novel hybrid mode enhance the QoS of P2P file sharing&releasing system? How much is the cost for renting cloud resuources, and what are the major influential factors of system cost? Conducting measurement study on real large-scale system to answer questions mentioned above, is important to understanding such systems. Secondly, cloud bandwidth provision of cloud-assisted file sharing&releasing system is scalable. How to match the demand and supply of bandwidth, in order to save cost while meeting QoS requirement, is also a critical problem to be solved. Thirdly, the novel services provide download appointment, and accelerate download tasks with both peer-assisted mode and cloud-assisted mode. Designing effieient network caching scheme based on such novel features, is a key issue for performance improvement and cost saving. The fourth, during the start-up phase of the releasing of popular file, the flash crowd of requests would lead to significant cloud traffic consumption. How to design cost saving strategy based on charging methods of cloud resources, also is an issue deserve to be studied.In addition, distributed architecture has been widely implemented by clouds which are supporting various popular services including file sharing&releasing. Studying cost minimizing problem of distributed cloud based on features of major services supported, in order to obtain optimal amount of distributed cloud computing-storage resources and their corresponding locations, is an issue with great academic value and pratical worth.Towards the problems mentioned above, this paper conducts many comprehensive and deep studies. The main research and innovation are summarized as follows:(1) Considering the lack of measurement study on cloud-assisted file distribution system, we conduct measurement and analysis on a large-scale representative, that is, QQ-Cyclone offline downloading system. Results not only show that two major shortages of traditional P2P systems, namely, can not guarantee file availability and can not assure high-speed download, are made up by cloud-assistance, but also reveal key factors which impact the cloud bandwidth consumption of download process.(2) To deal with the issue that exsiting server bandwidth allocation strategies are not suitable for the novel cloud-assisted system, we propose an adaptive cloud bandwidth allocation algorithm based on the major influential factors of cloud bandwidth consumption of download process which are obtained through practical measurement. Trace-driven experiments verify that our design significantly reduce the overhead of cloud bandwidth while satisfying QoS requirement.(3) Considering the new features like offering download appointment and providing both peer-acceleration and cloud-acceleration, we study caching problem for cloud-assisted file distribution system, and propose complementary caching which simultaneously deploys two different caching unities in individual ISP. P2P super node is responsible for caching scarce files in the ISP to increase capacity of local P2P network. Cache cloud extended node selects caching objects according to download appointments, and exclude files which are stored in P2P super node, in order to share the bandwidth requirement of cloud. A three-month practical measurement reveals that complementary cahing achieve the "three-win" goal, that is, alleviating bandwidth load for Cache cloud, enhancing download speed for users, and mitigating inter-ISP traffic.(4) To deal with the issue that too much cloud resources are consumed under flash crowd during the start-up phase of popular files’releasing, we firstly build the Fluid model of online game releasing process, and then derive the occupation time of cloud resources and consumption of cloud traffic. Throuth the analysis of typical releasing process, the real overhead of cloud resources is found to be much higher than iis theoretical bound when two files with different sizes are releasing simultaneously, because of the low utilization of users’uploading capacity. Therefore, we propose file chunk push scheme, which actively push a partition of the file with insufficient peer uploading capacity to "helper" with surplus uploading bandwidth. By solving the optimization problem aiming at minimizing the cloud resource consumption, the optimal push strategy is obtain. We finally evaluate the effectiveness of push strategy.(5) Considering cost minimization deployment of distributed cloud which hosts numerous popular services, we firstly build a generic expense model of cloud based on a abstracted system architecture, taking into account four major components of operating cost. Then, we formulate optimizations for three typical clouds supporting different main services, respectively. The optimal amount of edge cloud computing storage resources and their corresponding best locations are obtained by solving the optimizing problems. Finally, we discuss the results to reflect the practical significance.