Niklas Carlsson

Analysis of bittorrent-like protocols for on-demand stored media streaming

This paper develops analytic models that characterize the behavior of on-demand stored media content delivery using BitTorrent-like protocols. The models capture the effects of different piece selection policies, including Rarest-First and two variants of In-Order. Our models provide insight into transient and steady-state system behavior, and help explain the sluggishness of the system with strict In-Order streaming. We use the models to compare different retrieval policies across a wide range of system parameters, including peer arrival rate, upload/download bandwidth, and seed residence time. We also provide quantitative results on the startup delays and retrieval times for streaming media delivery. Our results provide insights into the optimal design of peer-to-peer networks for on-demand media streaming.

Characterizing Web-Based Video Sharing Workloads

Video sharing services that allow ordinary Web users to upload video clips of their choice and watch video clips uploaded by others have recently become very popular. This article identifies invariants in video sharing workloads, through comparison of the workload characteristics of four popular video sharing services. Our traces contain metadata on approximately 1.8 million videos which together have been viewed approximately 6 billion times. Using these traces, we study the similarities and differences in use of several Web 2.0 features such as ratings, comments, favorites, and propensity of uploading content. In general, we find that active contribution, such as video uploading and rating of videos, is much less prevalent than passive use. While uploaders in general are skewed with respect to the number of videos they upload, the fraction of multi-time uploaders is found to differ by a factor of two between two of the sites. The distributions of lifetime measures of video popularity are found to have heavy-tailed forms that are similar across the four sites. Finally, we consider implications for system design of the identified invariants. To gain further insight into caching in video sharing systems, and the relevance to caching of lifetime popularity measures, we gathered an additional dataset tracking views to a set of approximately 1.3 million videos from one of the services, over a twelve-week period. We find that lifetime popularity measures have some relevance for large cache (hot set) sizes (i.e., a hot set defined according to one of these measures is indeed relatively “hot”), but that this relevance substantially decreases as cache size decreases, owing to churn in video popularity.

Characterizing and modelling popularity of user-generated videos

This paper develops a framework for studying the popularity dynamics of user-generated videos, presents a characterization of the popularity dynamics, and proposes a model that captures the key properties of these dynamics. We illustrate the biases that may be introduced in the analysis for some choices of the sampling technique used for collecting data; however, sampling from recently-uploaded videos provides a dataset that is seemingly unbiased. Using a dataset that tracks the views to a sample of recently-uploaded YouTube videos over the first eight months of their lifetime, we study the popularity dynamics. We find that the relative popularities of the videos within our dataset are highly non-stationary, owing primarily to large differences in the required time since upload until peak popularity is finally achieved, and secondly to popularity oscillation. We propose a model that can accurately capture the popularity dynamics of collections of recently-uploaded videos as they age, including key measures such as hot set churn statistics, and the evolution of the viewing rate and total views distributions over time.

Evolution of an online social aggregation network: an empirical study

Many factors such as the tendency of individuals to develop relationships based on mutual acquaintances, proximity, common interests, or combinations thereof, are known to contribute toward evolution of social networks. In this paper, we analyze an evolving online social aggregator FriendFeed, which collates content generated by participating individuals on a variety of Web 2.0 services and allows easy dissemination of the aggregated content to other participants of the aggregator. Analyzing data collected between September 2008 and May 2009, we find that although preferential attachment captures the evolution of the network, its influence varies significantly based on how long ago a user joined the service. In particular, preferential attachment does not appear to apply to new entrants of the FriendFeed service. Analysis suggests that proximity bias plays an important role in link formation. We study the influence of common foci and find that individuals have a greater affinity toward those with similar interests.

A tale of the tails: Power-laws in internet measurements

Power-laws are ubiquitous in the Internet and its applications. This tutorial presents a review of power-laws with emphasis on observations from Internet measurements. First, we introduce power-laws and describe two commonly observed power-law distributions, the Pareto and Zipf distributions. Two frequently occurring terms associated with these distributions, specifically heavy tails and long tails, are also discussed. Second, the preferential attachment model, which is a widely used model for generating power-law graph structures, is reviewed. Subsequently, we present several examples of Internet workload properties that exhibit power-law behavior. Finally, we explore several implications of power-laws in computer networks. Using examples from past and present, we review how researchers have studied and exploited power-law properties. We observe that despite the challenges posed, power-laws have been effectively leveraged by researchers to improve the design and performance of Internet-based systems.

Server selection in large-scale video-on-demand systems

Video on demand, particularly with user-generated content, is emerging as one of the most bandwidth-intensive applications on the Internet. Owing to content control and other issues, some video-on-demand systems attempt to prevent downloading and peer-to-peer content delivery. Instead, such systems rely on server replication, such as via third-party content distribution networks, to support video streaming (or pseudostreaming) to their clients. A major issue with such systems is the cost of the required server resources. By synchronizing the video streams for clients that make closely spaced requests for the same video from the same server, server costs (such as for retrieval of the video data from disk) can be amortized over multiple requests. A fundamental trade-off then arises, however, with respect to server selection. Network delivery cost is minimized by selecting the nearest server, while server cost is minimized by directing closely spaced requests for the same video to a common server. This article compares classes of server selection policies within the context of a simple system model. We conclude that: (i) server selection using dynamic system state information (rather than only proximities and average loads) can yield large improvements in performance, (ii) deferring server selection for a request as late as possible (i.e., until just before streaming is to begin) can yield additional large improvements, and (iii) within the class of policies using dynamic state information and deferred selection, policies using only “local” (rather than global) request information are able to achieve most of the potential performance gains.

Characterizing Organizational Use of Web-Based Services: Methodology, Challenges, Observations, and Insights

Today’s Web provides many different functionalities, including communication, entertainment, social networking, and information retrieval. In this article, we analyze traces of HTTP activity from a large enterprise and from a large university to identify and characterize Web-based service usage. Our work provides an initial methodology for the analysis of Web-based services. While it is nontrivial to identify the classes, instances, and providers for each transaction, our results show that most of the traffic comes from a small subset of providers, which can be classified manually. Furthermore, we assess both qualitatively and quantitatively how the Web has evolved over the past decade, and discuss the implications of these changes.

Characterizing large-scale routing anomalies: a case study of the china telecom incident

China Telecoms hijack of approximately 50,000 IP prefixes in April 2010 highlights the potential for traffic interception on the Internet. Indeed, the sensitive nature of the hijacked prefixes, including US government agencies, garnered a great deal of attention and highlights the importance of being able to characterize such incidents after they occur. We use the China Telecom incident as a case study, to understand (1) what can be learned about large-scale routing anomalies using public data sets, and (2) what types of data should be collected to diagnose routing anomalies in the future. We develop a methodology for inferring which prefixes may be impacted by traffic interception using only control-plane data and validate our technique using data-plane traces. The key findings of our study of the China Telecom incident are: (1) The geographic distribution of announced prefixes is similar to the global distribution with a tendency towards prefixes registered in the Asia-Pacific region, (2) there is little evidence for subprefix hijacking which supports the hypothesis that this incident was likely a leak of existing routes, and (3) by preferring customer routes, providers inadvertently enabled interception of their customers traffic.

Using torrent inflation to efficiently serve the long tail in peer-assisted content delivery systems

A peer-assisted content delivery system uses the upload bandwidth of its clients to assist in delivery of popular content. In peer-assisted systems using a BitTorrent-like protocol, a content delivery server seeds the offered files, and active torrents form when multiple clients make closely-spaced requests for the same content. Scalability is achieved in the sense of being able to accommodate arbitrarily high request rates for individual files. Scalability with respect to the number of files, however, may be much more difficult to achieve, owing to a “long tail” of lukewarm or cold files for which the server may need to assume most or all of the delivery cost. This paper first addresses the question of how best to allocate server resources among multiple active torrents. We then propose new content delivery policies that use some of the available upload bandwidth from currently downloading clients to “inflate” torrents for files that would otherwise require substantial server bandwidth. Our performance results show that use of torrent inflation can substantially reduce download times, by more than 50% in some cases.

Helping Hand or Hidden Hurdle: Proxy-Assisted HTTP-Based Adaptive Streaming Performance

HTTP-based Adaptive Streaming (HAS) has become a widely-used video delivery technology. Use of HTTP enables relatively easy firewall/NAT traversal and content caching. While caching is an important aspect of HAS, there is not much public research on the performance impact proxies and their policies have on HAS. In this paper we build an experimental framework using open source Squid proxies and the most recent Open Source Media Framework (OSMF). A range of content-aware policies can be implemented in the proxies and tested, while the player software can be instrumented to measure performance as seen at the client. Using this framework, the paper makes three main contributions. First, we present a scenario-based performance evaluation of the latest version of the OSMF player. Second, we quantify the benefits using different proxy-assisted solutions, including basic best effort policies and more advanced content quality aware prefetching policies. Finally, we present and evaluate a cooperative framework in which clients and proxies share information to improve performance. In general, the bottleneck location and network conditions play central roles in which policy choices are most advantageous, as they significantly impact the relative performance differences between policy classes. We conclude that careful design and policy selection is important when trying to enhance HAS performance using proxy assistance.