Off-campus UNL users: To download campus access dissertations, please use the following link to log into our proxy server with your NU ID and password. When you are done browsing please remember to return to this page and log out.
Non-UNL users: Please talk to your librarian about requesting this dissertation through interlibrary loan.
Bandwidth Estimation in Cloud Networks
The bandwidth information is important for many network applications such as multimedia streaming applications and Transmission Control Protocols (TCP) to optimize their performance. However, most Internet providers do not provide their bandwidth information, and bandwidth can also change over time. Thus, many bandwidth estimation (BwEst) methods have been proposed to measure the network bandwidth by analyzing how packets are delivered in the network. Recent years have witnessed a rapid increase in public cloud adoption as the growing demands and cloud computing advantages. Cloud providers use virtual machine (VM) scheduling to virtualize physical CPUs and dynamically provide customers with various VM options. However, existing BwEst methods are designed for traditional wired and wireless networks and do not work well in clouds. Their filtering techniques work only for noises with durations in the order of microseconds or less. VM scheduling noises are longer (in the order of milliseconds or greater) and unpredictable making BwEst in clouds more challenging. This dissertation addresses VM scheduling challenges by rigorously analyzing current BwEst methods' accuracy in cloud networks. We are the first to identify the fundamental reason for their poor accuracy in clouds: VM scheduling. From these insights, we propose three solutions to improve BwEst’s performance for different applications and protocols in clouds. First, we propose a sequence-based BwEst method for cloud networks, which is fundamentally different from current time-based BwEst methods. Our proposed sequence-based methods are insensitive to VM scheduling, whereas current time-based methods are sensitive to VM scheduling. Second, we are the first to identify and study the fundamental reason for the performance of the emerging model-based TCP protocol in cloud networks and propose a new model to significantly improve their performance. Finally, for the first time, we identify and study the impact of VM scheduling on Multipath TCP, which is a TCP extension mainly for cellular networks and data centers, and propose a method to greatly improve its performance in cloud networks.
Ha, Phuong, "Bandwidth Estimation in Cloud Networks" (2021). ETD collection for University of Nebraska - Lincoln. AAI28414490.