Books by Alexandre Mateus
This research uses data collected from a university campus network via Deep Packet Inspection (DP... more This research uses data collected from a university campus network via Deep Packet Inspection (DPI) monitoring and from the largest public BitTorrent tracker to characterize the extent of unauthorized transfers of copyrighted content using Peer-to-Peer (P2P) and to evaluate the effectiveness and limitations of DPI in detection of such activity, both to provide a perspective of how much copyright infringement happens using P2P and to inform those seeking to deploy DPI technology.
Use of P2P and transfers of copyrighted content were widespread on campus. In Spring 2008, 40% of students living on campus were detected using a P2P protocol, 70% of which were observed attempting to transfer copyrighted material. In late 2010, we estimate that over 800 million copies of content were transferred globally using BitTorrent per day, with an estimated number of transferred songs 13.1 times greater than worldwide sales of songs, and estimated number of transferred movies 6.8 times greater than worldwide box-office sales and 16.4 times greater than U.S. DVD and Blu-ray sales. Most transfers were from a small number of very popular titles that were widely available for sale. We found no evidence that use of P2P to transfer content without violating copyright was common both on campus and global BitTorrent. This indicates that copyright law is violated frequently using P2P, and while we cannot quantify how P2P transfers translate to lost sales, it is reasonable to assume some sales are lost due to P2P.
Focusing on effectiveness of DPI, after a couple weeks of monitoring DPI found up to 80% of detected P2P users attempting to transfer copyrighted content. In the short term, DPI could be effective to assess which network users transfer copyrighted content using P2P given some weeks of monitoring. However, limitations such as not being able to detect users of encrypted P2P can reduce DPIʼs effectiveness in the long term. Using behavioral classifiers that we implemented and that can detect encrypted BitTorrent from traffic summaries, we found students shifting from unencrypted to encrypted BitTorrent in the 2007-2008 academic year. If this trend continues, effectiveness of DPI for enforcement can be significantly hindered.
Papers by Alexandre Mateus
Telecommunications Policy, Sep 17, 2013
The effectiveness of many proposed policies regarding both online copyright protection and networ... more The effectiveness of many proposed policies regarding both online copyright protection and network neutrality depend on the extent to which it is technically possible to detect peer-to-peer file sharing (P2P), the transfer of copyrighted files, or both. There are many detection approaches, some performed primarily by network operators and some by application-layer agents. This paper describes capabilities, limitations, privacy issues, and policy implications of detection technologies and their countermeasures, in part through quantitative analysis of empirical data. Different approaches are better for different purposes. Network operators are well-positioned to estimate how widespread copyright violations are, but application-layer detection from outside entities has important advantages when the purpose is punishment. Detection is also imperfect, so policies should require more transparency regarding how it is done than we see today. It is shown that, although network operators may not detect every transfer, and they typically miss more video than audio, they can identify most individuals who share copyrighted files via P2P after several weeks of monitoring provided that traffic is unencrypted, which is useful for some purposes. However, it is also shown that encryption is already in use, and it effectively prevents network operators from detecting transfers of copyrighted content. Thus, if network operators are held responsible for monitoring illegal file sharing, there is a tension between using detection to identify violators of copyright law for punishment, which may motivate even greater use of encryption, and using detection for other purposes such as creating fair compensation schemes for copyright-holders, warning users that they may be violating copyright law, or allocating network resources. Alternatively, there are forms of detection that are not evaded through encryption, and application-layer agents rather than network operators are primarily responsible for these. These copyright policy issues are intertwined with network neutrality policy in subtle ways. Network neutrality rules do not protect illegal transfers of copyrighted content, but if network operators are responsible for enforcement (as in “graduated response”) then regulators must determine when it is reasonable to terminate or degrade service based on allegations of copyright violation given the limitations of detection technology to prove those allegations. Allegations of copyright violation should be considered invalid unless they are accompanied with information about how detection was performed and an opportunity for rebuttal. Such transparency has been routinely lacking in both laws and industry agreements.
This paper presents the most accurate empirical study to date to characterize and quantify the am... more This paper presents the most accurate empirical study to date to characterize and quantify the amount of content of various types that is transferred worldwide using BitTorrent, the dominant peer-to-peer (P2P) file sharing application. Using data we collected from the largest public BitTorrent tracker over 106 days between August 2010 and February 2011 and a new methodology, we find that for some content types, the number of copies transferred is an order of magnitude greater than the number sold through legal channels. For example, we estimate that 10.7 songs were transferred using BitTorrent for every song sold, 3.6 movies were transferred using BitTorrent for every legal sale or rental of a DVD or Blu-ray, and 227 movies were transferred using BitTorrent for every paid download. We also find that the vast majority of music and video content transferred using BitTorrent is copyrighted, as demonstrated both by the swarm metadata we observed, and the fact that only 0.55% of the transfers were of files indexed by websites that specialize in content that can be transferred legally. Thus, we conclude that BitTorrent transfers result in hundreds of millions of copyright violations worldwide per day, and that copyright holders fail to realize significant revenues as a result. Movies are the type of content most supplied and most transferred in BitTorrent (shared in 38.7% of swarms and accounting for 26.1% of transfers). Songs and software, despite being shared in small percentages of swarms (4.5% and 7.2% of swarms respectively), rank 2nd and 3rd in terms of transfers (with 20.4% and 16.8% of transfers respectively). This shows the limitations of past studies that estimated the economic impact of P2P by looking at which content is available rather than trying to measure the number of actual transfers. Surprisingly, most of the copies transferred using BitTorrent come from a small number of extremely popular titles; 37 song titles account for half of all songs transferred, and 117 movies account half of all movie transfers. Thus, for a global marketplace, the importance of the “long tail” of less popular content is smaller than we and others have observed in more localized studies. In general, the content that is popular in legal channels is also popular with BitTorrent, but we observe some important differences. For example, we find that content that is popular among teenagers is more likely to be disproportionally represented in BitTorrent transfers as compared to content that appeals to an older audience.
This article presents findings from a large-scale quantitative assessment of online exchanges of ... more This article presents findings from a large-scale quantitative assessment of online exchanges of copyrighted material on a college campus based on network data collected using deep packet inspection (DPI). We find that use of Peer-to-Peer (P2P) for the transfer of copyrighted content is widespread on campus, although observed P2P is declining. In a month-long monitoring period in Spring 2008, at least 40% of students living on campus were observed engaging in P2P, 70% of those were detected attempting to transfer copyrighted content, and each of the latter was observed transferring copyrighted titles at an average rate of 4 titles per day. Nevertheless, from Spring 2007 to Spring 2008, the daily percentage of detected P2P users fell 10%, and the daily percentage of users observed attempting to transfer copyrighted content out of those detected doing P2P fell 20%. These changes could be the result of decreasing use of P2P, or increasing use of encrypted P2P to evade detection. We also find that, given a couple weeks or more, current DPI technology identifies most users attempting to transfer copyrighted material, out of users whose P2P traffic it can detect. This shows that even if DPI does not detect every transfer of copyrighted material, it can effectively identify individuals who make these transfers, provided they do not use encryption. However, detection of copyrighted content is less accurate for video than for audio, so it may take far longer to identify individuals who use P2P to transfer copyrighted video but not copyrighted audio. Finally, to shed light on the impact of P2P on sales of content, we find that 22% of P2P users also purchase content from the iTunes Store (iTS), each buying on average about as much content as non- P2P users who purchase from the iTunes Store. This refutes the hypothesis that all P2P users view the ability to obtain free content from P2P as a complete substitute for paying for content. On the other hand, we also find that among iTs users, those who use P2P are somewhat more likely than those do not use P2P to access iTs only for the free samples.
This article presents findings from the first large-scale quantitative assessment of Peer-to-Peer... more This article presents findings from the first large-scale quantitative assessment of Peer-to-Peer (P2P) exchanges of copyrighted material on a college campus based on actual observation. Through passive monitoring and deep packet inspection (DPI), we assess the extent to which P2P is used to transfer copyrighted material. We also characterize the demographics of P2P users, the relative popularity of the material, and how the burden on the campus network varies over time. We found that at least 51% of students living on campus engaged in P2P, at least 42% attempted to transfer copyrighted material, and the mean number of copyrighted media titles whose transfer is attempted per week was at least 6 per monitored student. Some students use P2P legally, e.g. to transfer Linux software or non-copyrighted adult material, but we found no evidence that large numbers of students use P2P for these legal purposes and not to transfer copyrighted material. Students of all genders, ages, classes and majors engaged in file sharing, to the extent that demographics were not helpful in identifying likely file-sharers so as to target interventions. This study also provides lessons for those who would use DPI technology to reduce illegal use of P2P. If given enough weeks to observe, current technology is effective at identifying users who attempt to transfer copyrighted material, provided that their traffic is identifiable as P2P. Thus, DPI can be used to estimate the extent of piracy, and to notify individuals who may be violating copyright law. However, encryption is available and can be easily activated in most P2P clients. Once turned on, encryption prevents DPI from detecting whether transferred material is copyrighted, rendering it ineffective. If DPI is used for copyright enforcement that includes imposition of penalties, then P2P users or P2P developers may have the incentive to use encryption as a way of evading detection.
The Internet is provisioned selfishly by Internet Service Providers (ISPs), which results in econ... more The Internet is provisioned selfishly by Internet Service Providers (ISPs), which results in economic inefficiency. We shed light on the sources of such inefficiency by means of a multi-commodity flow model of how ISPs interconnect and provision communications infrastructure to deliver an ex- ogenous amount of demand. The outcome of the game ISPs play is a set of equilibrium networks sustained by intercon- nection prices computed assuming that transit markets are competitive. Our theoretical results show that equilibrium networks exist under mild assumptions. We also show that on the grounds of cost efficiency, ISPs have no incentive to split traffic flows thus choosing a single path for all the traffic flowing from a specific origin to a specific destina- tion. Furthermore, we show that network provisioning inef- ficiency follows an inverted U-shaped curve as a function of the level of economies of scale. Simulation over small-sized networks confirms our theoretical results. Simulation also shows that ISPs enjoying higher levels of economies of scale in network provisioning become long-haul transit providers and regional ISPs depend on them to deliver their traffic. Using such a transit provider, who can benefit largely from economies of scale, reduces overall provisioning costs close to optimal.
Uploads
Books by Alexandre Mateus
Use of P2P and transfers of copyrighted content were widespread on campus. In Spring 2008, 40% of students living on campus were detected using a P2P protocol, 70% of which were observed attempting to transfer copyrighted material. In late 2010, we estimate that over 800 million copies of content were transferred globally using BitTorrent per day, with an estimated number of transferred songs 13.1 times greater than worldwide sales of songs, and estimated number of transferred movies 6.8 times greater than worldwide box-office sales and 16.4 times greater than U.S. DVD and Blu-ray sales. Most transfers were from a small number of very popular titles that were widely available for sale. We found no evidence that use of P2P to transfer content without violating copyright was common both on campus and global BitTorrent. This indicates that copyright law is violated frequently using P2P, and while we cannot quantify how P2P transfers translate to lost sales, it is reasonable to assume some sales are lost due to P2P.
Focusing on effectiveness of DPI, after a couple weeks of monitoring DPI found up to 80% of detected P2P users attempting to transfer copyrighted content. In the short term, DPI could be effective to assess which network users transfer copyrighted content using P2P given some weeks of monitoring. However, limitations such as not being able to detect users of encrypted P2P can reduce DPIʼs effectiveness in the long term. Using behavioral classifiers that we implemented and that can detect encrypted BitTorrent from traffic summaries, we found students shifting from unencrypted to encrypted BitTorrent in the 2007-2008 academic year. If this trend continues, effectiveness of DPI for enforcement can be significantly hindered.
Papers by Alexandre Mateus
Use of P2P and transfers of copyrighted content were widespread on campus. In Spring 2008, 40% of students living on campus were detected using a P2P protocol, 70% of which were observed attempting to transfer copyrighted material. In late 2010, we estimate that over 800 million copies of content were transferred globally using BitTorrent per day, with an estimated number of transferred songs 13.1 times greater than worldwide sales of songs, and estimated number of transferred movies 6.8 times greater than worldwide box-office sales and 16.4 times greater than U.S. DVD and Blu-ray sales. Most transfers were from a small number of very popular titles that were widely available for sale. We found no evidence that use of P2P to transfer content without violating copyright was common both on campus and global BitTorrent. This indicates that copyright law is violated frequently using P2P, and while we cannot quantify how P2P transfers translate to lost sales, it is reasonable to assume some sales are lost due to P2P.
Focusing on effectiveness of DPI, after a couple weeks of monitoring DPI found up to 80% of detected P2P users attempting to transfer copyrighted content. In the short term, DPI could be effective to assess which network users transfer copyrighted content using P2P given some weeks of monitoring. However, limitations such as not being able to detect users of encrypted P2P can reduce DPIʼs effectiveness in the long term. Using behavioral classifiers that we implemented and that can detect encrypted BitTorrent from traffic summaries, we found students shifting from unencrypted to encrypted BitTorrent in the 2007-2008 academic year. If this trend continues, effectiveness of DPI for enforcement can be significantly hindered.