Live streaming routinely relies on the Hypertext Transfer Protocol (HTTP) and content delivery networks (CDNs) to scalably disseminate videos to diverse clients. A bitrate ladder refers to a list of bitrate-resolution pairs, or representations, used for encoding a video. A promising trend in HTTP-based video streaming is to adapt not only the client's representation choice but also the bitrate ladder during the streaming session. This paper examines the problem of multi-live streaming, where an encoding service coordinates CDN-assisted bitrate ladder adaptation for multiple live streams delivered to heterogeneous clients in different zones via CDN edge servers. We design ALPHAS, a practical and scalable system for multi-live streaming that accounts for CDNs' bandwidth constraints and encoders' computational capabilities and also supports stream prioritization. ALPHAS, aware of both video content and streaming context, seamlessly integrates with the end-to-end streaming pipeline and operates in real time transparently to clients and encoding algorithms. We develop a cloud-based ALPHAS implementation and evaluate it through extensive real-world and trace-driven experiments against four prominent baseline approaches that encode each stream independently. The evaluation shows that ALPHAS outperforms the baselines, improving quality of experience, end-to-end latency, and per-stream processing by up to 23%, 21 %, and 49%, respectively.

Remaining a dominant force in Internet traffic, video streaming captivates end users, service providers, and researchers. This article takes a pragmatic approach to reviewing recent advances in the field by focusing on the prevalent streaming paradigm that involves delivering long-form two-dimensional videos over the best-effort Internet with client-side adaptive bitrate (ABR) algorithms and assistance from content delivery networks (CDNs). To enhance accessibility, we supplement the survey with tutorial material. Unlike existing surveys that offer fragmented views, our work provides a holistic perspective on the entire end-to-end streaming pipeline, from video capture by a camera-equipped device to playback by the end user. Our novel perspective covers the ingestion, processing, and distribution stages of the pipeline and addresses key challenges such as video compression, upload, transcoding, ABR algorithms, CDN support, and quality of experience. We review over 200 papers and classify streaming designs by problem-solving methodology, whether based on intuition, theory, or machine learning. The survey further refines these methodology-based categories and characterizes each design by additional traits such as compatible codecs. We connect the reviewed research to real-world applications by discussing the practices of commercial streaming platforms. Finally, the survey highlights prominent current trends and outlines future directions in video streaming.

Quality of experience (QoE) becomes both the holy grail and a free-for-all in adaptive bitrate (ABR) video streaming. On the one hand, the design, operation, and evaluation of ABR algorithms increasingly rely on QoE. On the other hand, QoE frequently receives only cursory attention in this supporting role, with many of its important aspects treated with insufficient care. As a complex subjective notion, QoE is directly measurable through subjective tests, which incur evident overhead. While an objective QoE model represents a scalable automated means for QoE assessment, QoE models proliferate without consensus on their goodness due to numerous influence factors, construction methods, and usages. The model proliferation creates a false impression that proposing a new QoE model without a proper validation is acceptable. Because the multifaceted QoE problem involves separable and often separated tasks of test conducting, model building, and model using, this separation of concerns causes additional complications. By leveraging two large real datasets of individual QoE perception, this paper reviews the status quo in QoE, identifies various pitfalls, and offers guidelines for test conducting, model building, and model using, so as to foster high standards in future work on QoE in ABR video streaming.

While ISPs (Internet service providers) strive to improve QoE (quality of experience) for end users, end-to-end traffic encryption by OTT (over-the-top) providers undermines independent inference of QoE by an ISP. Due to the economic and technological complexity of the modern Internet, ISP-side QoE inference based on OTT assistance or out-of-band signaling sees low adoption. This paper presents IQN (in-band quality notification), a novel mechanism for signaling QoE impairments from an automated agent on the end-user device to the server-to-client ISP responsible for QoE-impairing congestion. Compatible with multi-ISP paths, asymmetric routing, and other Internet realities, IQN does not require OTT support and induces the OTT server to emit distinctive packet patterns that encode QoE information, enabling ISPs to infer QoE by monitoring these patterns in network traffic. We develop a prototype system, YouStall, which applies IQN signaling to ISP-side inference of YouTube stalls. Cloud-based experiments with YouStall on YouTube Live streams validate IQN's feasibility and effectiveness, demonstrating its potential for accurate user-assisted ISP-side QoE inference from encrypted traffic in real Internet environments.

Live streaming of segmented videos over the Hypertext Transfer Protocol (HTTP) is increasingly popular and serves heterogeneous clients by offering each segment in multiple representations. A bitrate ladder expresses this choice as a list of bitrate-resolution pairs. Whereas existing solutions for HTTP-based live streaming use a static bitrate ladder, the fixed ladders struggle to appropriately accommodate the dynamics in the video content and network-conditioned client capabilities. This paper proposes ARTEMIS as a practical scalable alternative that dynamically configures the bitrate ladder depending on the content complexity, network conditions, and clients' statistics. ARTEMIS seamlessly integrates with the end-to-end streaming pipeline and operates transparently to video encoders and clients. We develop a cloud-based implementation of ARTEMIS and conduct extensive real-world and trace-driven experiments. The experimental comparison vs. existing prominent bitrate ladders demonstrates that live streaming with ARTEMIS outperforms all baseline solutions, reduces encoding computation by 25%, end-to-end latency by 18%, and increases the quality of experience by 11%.

Quality of Experience (QoE) and QoE models are of an increasing importance to networked systems. The traditional QoE modeling for video streaming applications builds a one-size-fits-all QoE model that underserves atypical viewers who perceive QoE differently. To address the problem of atypical viewers, this paper proposes iQoE (individualized QoE), a method that employs explicit, expressible, and actionable feedback from a viewer to construct a personalized QoE model for this viewer. The iterative iQoE design exercises active learning and combines a novel sampler with a modeler. The chief emphasis of our paper is on making iQoE sample-efficient and accurate. By leveraging the Microworkers crowdsourcing platform, we conduct studies with 120 subjects who provide 14,400 individual scores. According to the subjective studies, a session of about 22 minutes empowers a viewer to construct a personalized QoE model that, compared to the best of the 10 baseline models, delivers the average accuracy improvement of at least 42% for all viewers and at least 85% for the atypical viewers. The large-scale simulations based on a new technique of synthetic profiling expand the evaluation scope by exploring iQoE design choices, parameter sensitivity, and generalizability.

In-band real-time telemetry is a promising direction for management of modern programmable networks. While network noise in the form of packet reordering and loss affects inband collection of distributed state, there is a need to compute telemetry functions on the collected state correctly despite the network noise. To address this common need, we propose TeleNoise that equips each packet with few sync bits and offers primitives of group affiliation and group completion to support noise-resilient computation of per-group telemetry functions. This paper gives real-world examples of such functions, elaborates on the role of TeleNoise in a modular in-band telemetry architecture, and presents algorithms for the two TeleNoise primitives. We derive analytical guarantees on correctness and performance of the algorithms and report a trace-driven evaluation that corroborates the effective low-overhead profile of TeleNoise, e.g., the assuredly correct operation and at most 1.6 packets of the average measurement lag for 12-packet groups and 3 sync bits.

While secure efficient operation of computer networks requires cost-effective line-rate packet classification, network programmability strengthens this need. A promising approach is to transform a packet classifier to a semantically equivalent representation that supports more effective classification. This paper explores transformation of ternary classifiers to equivalent prefix representations so that classification can benefit from efficient Longest Prefix Match solutions. We propose the property of prefix disjointness and design PREDICAT, a method that leverages this new property in combination with a variety of existing techniques to convert an arbitrary ternary classifier to an equivalent prefix representation. The paper analyzes prefix disjointness and evaluates PREDICAT against state-of-the-art transformation alternatives on a packet classification benchmark in regard to the number of lookups. The evaluation shows that PREDICAT outperforms a ternary-to-binary method by up to an order of magnitude, improves on another ternary-to-prefix solution by up to a factor of 5, and performs similarly to a ternary-to-ternary approach that requires costly power-hungry Ternary Content-Addressable Memories to efficiently handle the resulting ternary representation.

Presents the introductory welcome message from the conference proceedings. May include the conference officers' congratulations to all involved with the conference event and publication of the proceedings record.

This editorial note aims to first inform the SIGCOMM community on the reviewing process in place currently at CCR, and second, share our plans to make CCR a more open and welcoming venue by making changes to the review process, adding more value to the SIGCOMM community.

Content delivery networks (CDNs) distribute much of the Internet content by caching and serving the objects requested by users. A major goal of a CDN is to maximize the hit rates of its caches, thereby enabling faster content downloads to the users. Content caching involves two components: an admission algorithm to decide whether to cache an object and an eviction algorithm to determine which object to evict from the cache when it is full. In this paper, we focus on cache admission and propose a novel algorithm called RL-Cache that uses model-free reinforcement learning (RL) to decide whether or not to admit a requested object into the CDN's cache. Unlike prior approaches that use a small set of criteria for decision making, RL-Cache weights a large set of features that include the object size, recency, and frequency of access. We develop a publicly available implementation of RL-Cache and perform an evaluation using production traces for the image, video, and web traffic classes from Akamai's CDN. The evaluation shows that RL-Cache improves the hit rate in comparison with the state of the art and imposes only a modest resource overhead on the CDN servers. Further, RL-Cache is robust enough that it can be trained in one location and executed on request traces of the same or different traffic classes in other locations of the same geographic region. The paper also reports extensive analyses of the RL-Cache sensitivity to its features and hyperparameter values. The analyses validate the made design choices and reveal interesting insights into the RL-Cache behavior.

Unrelenting traffic growth, device heterogeneity, and load unevenness create scalability challenges for traffic monitoring. In this paper, we propose Robust Distributed Computation (RoDiC), a new approach that addresses these challenges by shifting a portion of the monitoring-task execution from an overloaded network element to another element that has spare resources. Moving the entire execution of the task away from the overloaded element might be infeasible because execution on multiple elements is inherent in the task or requires at least partial participation by the designated overloaded element. Furthermore, distributed execution of a stateful task has to be resilient to network noise in the form of packet reordering and loss. The RoDiC approach relies on two main principles of packet grouping and state overlap to support exact robust distributed monitoring of traffic flows under network noise. RoDiC uses an open-loop paradigm that does not add any control packets, communicates flow state in-band by appending few control bits to packets of monitored flows, and keeps measurement latency low. We apply RoDiC to the problem of flow-size computation and discuss how to instantiate our general technique for real-time packet-loss telemetry. The paper develops robust algorithms, proves their correctness and performance properties, and reports an evaluation driven by realistic traffic traces. The RoDiC algorithms successfully distribute the monitoring-task load while keeping the memory and computation overhead low.

New protocols for Internet inter-domain routing struggle to get widely adopted. Because the Internet consists of more than 50,000 autonomous systems (ASes), deployment of a new routing protocol has to be incremental. In this work, we study such incremental deployment. We first formulate the routing problem in regard to a metric of routing cost. Then, the paper proposes and rigorously defines a statistical notion of protocol ignorance that quantifies the inability of a routing protocol to accurately determine routing prices with respect to the metric of interest. The proposed protocol-ignorance model of a routing protocol is fairly generic and can be applied to routing in both inter-domain and intra-domain settings, as well as to transportation and other types of networks. Our model of protocol deployment makes our study specific to Internet inter-domain routing. Through a combination of mathematical analysis and simulation, we demonstrate that the benefits from adopting a new inter-domain protocol accumulate smoothly during its incremental deployment. In particular, the simulation shows that decreasing the routing price by 25% requires between 43% and 53% of all nodes to adopt the new protocol. Our findings elucidate the deployment struggle of new inter-domain routing protocols and indicate that wide deployment of such a protocol necessitates involving a large number of relevant ASes into a coordinated effort to adopt the new protocol.

Network devices need packet counters for a variety of applications. For a large number of concurrent flows, on-chip memories can be too small to support a separate counter per flow. While a single network element might struggle to implement flow accounting on its own, in this work we study alternatives leveraging underutilized resources elsewhere in the network and implement flow accounting on multiple network devices. This paper takes the first step towards understanding the design principles for robust network-wide accounting with lossy unidirectional channels without feedback.

Faced with the unrelenting traffic growth and rising costs, access ISPs (Internet Service Providers) realize that their traditional revenue model of flat subscription fees is unsustainable. Regulatory concerns, e.g., about content-specific charging, constrain the ISPs' search for new revenue sources. This paper analyzes a revenue model where an access ISP acts as a publisher of ads to users who explicitly opt for an ad-sponsored access plan of the ISP. First, we conduct a survey showing a substantial interest of users in ad-sponsored Internet access. Then, we mathematically characterize the advertising revenue model and asses its economic feasibility based on real data from two access ISPs and an ad publisher. While the ad revenues are tangible, they do not completely cover the costs of the access ISPs. Even in relative terms, a larger access ISP benefits more from the advertising, with the ad revenues covering up to 50% of the ISP's capital expenditure. Complementing the subscription fees, the access ISPs can leverage the ad revenues to meaningfully incentivize the users with better Internet connectivity, such as 6-9 Mbps in extra data rates or 12-20 GB in extra data caps for the two considered ISPs.

Ads (advertisements) have become a common source of revenues for websites and transformed the web hosting ecosystem. This paper reports an extensive measurement-based study of web hosting with an explicit focus on differences in the hosting of ads versus regular contents. Using a VPN (Virtual Private Network) service and 22,040 open recursive DNS (Domain Name System) resolvers, we collect contents of top country-specific websites in 52 countries around the globe and characterize the hosting infrastructures. While we observe that ads employ more servers for broader load distribution, replication is local for ads and global for regular contents. Our results clearly show that transit ASes (Autonomous Systems), including the tier-1 networks, diversify their economic roles and prominently provide web hosting. A small number of ASes dominate heavily in the byte volume of hosted contents, with the top hosts being different for ads and regular contents. Compared to ASes and organizations, the distribution of hosting countries is even more heavily skewed, with the USA consistently taking the overwhelmingly dominant position. While ads have shorter response times, their download times are longer because websites are developed to serve requests for regular contents with a higher priority.

A computer network serves distributed applications by communicating messages between their remote ends. Many such applications desire minimal delay for their messages. Beside this efficiency objective, allocation of the network capacity is also subject to the fairness constraint of not shutting off communication for any individual message. Processor Sharing (PS) is a de facto standard of fairness but provides significantly higher average delay than Shortest Remaining Processing Time (SRPT), which is an optimally efficient but unfair algorithm. In this paper, we explore efficient fair algorithms for message communication where fairness means that no message is delivered later than under PS. First, we introduce a slack system to characterize fair algorithms completely and develop efficient fair algorithms called Pessimistic Fair Sojourn Protocol (PFSP), Optimistic Fair Sojourn Protocol (OFSP), and Shortest Fair Sojourn (SFS). Then, we prove that a fair online algorithm does not assure minimal average delay attainable with fairness. Our analysis also reveals lower bounds on worst-case inefficiency of fair algorithms. We conduct extensive simulations for various distributions of message sizes and arrival times. During either temporary overload or steady-state operation, SFS and other newly proposed fair algorithms support SRPT-like efficiency and consistently provide much smaller average delay than PS.

In this paper, we revisit the question of how much buffer an IP router should allocate for its output link. For a long time, the intuitive answer of setting the buffer size to the bitrate-delay product has been widely regarded as reasonable. Recent studies of interaction between queueing at IP routers and TCP congestion control proposed alternative answers. First, we expose and explain contradictions between existing guidelines for link buffer sizing. Then, we argue that the problem of link buffer sizing needs a different formulation. In particular, the chosen buffer size should accommodate not only common versions of TCP but also UDP traffic. Besides, our new formulation of the problem contains an explicit constraint of not engaging IP routers in any additional signaling. We conclude the paper by outlining a promising direction for solving the reformulated problem. Link Buffer Sizing: a New Look at the Old Problem Sergey Gorinsky Anshul Kantawala Jonathan Turner Applied Research Laboratory Department of Computer Science and Engineering Washington University in St. Louis St. Louis, MO 63130-4899, USA gorinsky,anshul,jst @arl.wustl.edu

Direct measurements of the real world are typically expensive and time-consuming. Recent studies have revealed the promising quick alternative where real-world characteristics are indirectly estimated based on data readily available in the Internet. Such methods for indirect virtual-world estimation of real-world properties rely on the implicit hypothesis that the estimation accuracy depends on how extensively the Internet penetrates into the real world. Our paper makes the first step towards validating this hypothesis. We adopt a simple statistical model to quantify the relationship between an index of consumer prices and online searches for respective categories of goods and services across 19 countries. We also show how the strength of this relationship depends on Internet-penetration and other socioeconomic variables.