Recent research has proposed that Content Delivery Networks (CDNs) can use better bandwidth allocation to improve video streaming services through congested links. Because CDNs are usually not located at the bottleneck link, shared bottleneck (co-bottleneck) detection on the video servers is necessary for joint flow shaping and the Quality of Experience (QoE) improvements. However, co-bottleneck detection is challenging in such environments due to the large number of flows, possible network topologies, and traffic patterns. Current detectors fail to balance detection accuracy, speed and overhead, and suffer performance degradation in the scale of thousands of flows on each video server. We propose FlowBot, a novel model-based passive co-bottleneck detector designed for deployment on a video server. FlowBot uses Siamese model to learn flow representations, and combines the training procedure with its clustering algorithm to continue to provide strong performance with up to thousands of flows. Our evaluations show that FlowBot can achieve consistently high accuracy (over 70% F1 with around 90% precision) in most tested scenarios, while maintaining a short detection delay of 3 s and overhead similar to the fastest benchmark algorithms.

Internet content providers often deliver content through bandwidth bottlenecks that are out of their control. Thus, despite often having massively over-provisioned upstream servers, the content providers still cannot control the end-to-end user experience. This paper explores remote traffic shaping, allowing the content provider to allocate its share of a remote bottleneck link across its users using a metric other than TCP fairness, while remaining TCP-friendly to cross traffic on the bottleneck link. To evaluate this approach, we designed FlowTele, the first system that shapes outbound traffic on an Internet-scale network to optimize provider-selected metrics, using source control with neither in-network support nor special client support. Our extensive evaluations over the Internet show that by strategically reallocating bandwidth among provider-owned co-bottlenecked flows, FlowTele improves the provider's total revenue by roughly 20%--30% in various network settings, compared with both (i) status quo TCP fairshare and (ii) recent practice by content providers that proactively throttles video quality during the COVID-19 pandemic, while being TCP-friendly to cross-traffic. Besides revenue, we also study other metrics, such as QoE fairness, that a content provider may wish to optimize using FlowTele.

In a vehicular edge computing (VEC) system, some vehicles with surplus computing resources can provide computation task offloading opportunities for other vehicles or pedestrians. However, the vehicular network is highly dynamic, with fast varying channel states and computation loads. These dynamics are difficult to model or to predict, but they have a major impact on the quality of service (QoS) of task offloading, including delay performance and service reliability. Meanwhile, the computing resources in VEC are often redundant due to the high density of vehicles. To improve the QoS of VEC and exploit the abundant computing resources on vehicles, we propose a learning-based task replication algorithm (LTRA) based on combinatorial multi-armed bandit (CMAB) theory, in order to minimize the average offloading delay. LTRA enables multiple vehicles to process the replicas of the same task simultaneously, and vehicles that require computing services can learn the delay performance of other vehicles while offloading tasks. We take the occurrence time of vehicles into consideration, and redesign the utility function of existing CMAB algorithm, so that LTRA can adapt to the time varying network topology of VEC. We use a realistic highway scenario to evaluate the delay performance and service reliability of LTRA through simulations, and show that compared with single task offloading, LTRA can improve the task completion ratio with deadline 0.6s from 80% to 98%.