Study on net-neutrality regulation
Dieses Dokument ist Teil der Anfrage „Erweiterte Anfragen zur Netzneutralität“
Study on net-neutrality regulation | 28 4.3 USA: Netalyzr Netalyzr was created by the ICSI, with funding from the US National Science Foundation (NSF), the Department of Homeland Security (DHS) Science and Technology Directorate’s Cyber Security Division, and donations from Amazon, Comcast, Google and Heise Netze. Netalyzr was joint winner of a 2011 FCC competition for innovative research and useful apps that further the understanding of Internet connectivity and network science. It is a Java applet that runs on the end user’s device and communicates with ICSI servers to measure performance through the network. Netalyzr runs a series of tests that collectively take a few minutes. The tests verify a wide range of properties of the end-user machine and the network to which it is connected, including: • upload and download bandwidth • network latency • TCP connection setup latency • size of the TCP uplink and downlink buffer • ability to directly access a wide range of TCP port numbers • presence or absence of a hidden proxy server • various HTTP (i.e. the language of the web) characteristics • various Domain Name System (DNS) characteristics, including whether each DNS resolver is responding to all types of requests • whether there is IPv6 connectivity (the newer version of the IP), and whether the browser and the DNS support IPv6. These tests can be illuminating, and can identify certain kinds of traffic degradation. It may be possible to identify some deviations from net neutrality using Netalyzr; however, there are other deviations that it is unlikely to uncover. 70 Although Netalyzr was joint winner of an FCC competition, the FCC makes no systematic use of this tool. The results could, however, form the basis for a complaint to the FCC. 4.4 USA: Measuring Broadband America programme Measuring Broadband America is the most notable active net-neutrality monitoring programme underway in the USA. The programme primarily reports on the speed, latency and packet loss that an end user receives through the participating ISP’s network. 70 Consider, for example, the TCP RESET packets that were interjected by Comcast in the BitTorrent incident. These were detected only because an end user was curious, had an Ethernet packet trace capability available, and knew enough to be able to interpret the results. It is difficult to see how a general tool like Netalyzr could identify such subtle blockages. Ref: 2009152-254 .
Study on net-neutrality regulation | 29 As the FCC explains, 71 the “[…] measurements that provide the underlying data in [the Measuring Broadband America] Report rely both on measurement clients and measurement servers. The measurement clients reside in the homes of 4,281 panellists who receive service by the 13 participating ISPs. The participating ISPs collectively account for over 80% of U.S. residential broadband Internet connections. The panellists closely match the overall state and region statistics of Internet access connections in the United States [based on the reporting data that network operators provide to the FCC]. […] Our methodology focuses on the performance of each participating ISP’s network. The metrics discussed in this Report are derived from traffic flowing between a measurement client (located within the modem or router within a panellist’s home) and a measurement server. For each panellist, the tests use the measurement server for which the latency between the measurement client and server is the lowest value. As a result, the metrics measure performance along a specific path within each ISP’s network, through a point of interconnection between the ISP’s network and the network on which the chosen measurement server resides.” The statistics captured are: • Download speed: measured for each client at 5-second intervals within a 30-second time interval, every 2 hours • Upload speed: measured for each client at 5-second intervals within a 30-second time interval, every 2 hours • Latency and packet loss: the round-trip times for approximately 2000 packets per hour, sent at randomly distributed times • Web browsing: the total time to request and receive web pages from nine popular websites (including the text and images), measured every hour. 72 Since 2011, the FCC has published annual reports on the Measuring Broadband America programme, including information on each of the indicators that has been measured. The FCC’s summarisation across the ISPs changed in 2016 in comparison with previous years, with a shift to the use of medians, quantiles and weighted averages, rather than simple averages. 73 In terms of measured download speed, the annual reports provide data for each of the 13 monitored ISPs, across a number of speed brackets. In each case, data is presented on the fraction of users who receive more than 95% of the advertised speed, those who receive between 80% and 95% of the advertised speed, and those who receive less than 80% of the advertised speed (see Figure 4). 71 US FCC (2017), 2016 Measuring Broadband America: Fixed Broadband Report: A Report on Consumer Fixed Broadband Performance in the United States, p.25. 72 Ibid, p.27, which provides additional detail. 73 Ibid, Appendix A. Ref: 2009152-254 .
Study on net-neutrality regulation | 30 Figure 4: ISP comparison – Total – weekday peak hours, 2016 [Source: Measuring Broadband America programme online data, 2016 74] 100% 90% 80% 70% Percentage of Panelists 60% 50% 40% 30% 20% 10% 0% CenturyLink Windstream Cox Optimum AT&T - IPBB ViaSat/Exede Charter Hughes AT&T - DSL Frontier Frontier Mediacom TWC Verizon Comcast Verizon DSL Cable Fiber Satellite >95% of advertised 80% - <95% of advertised <80% of advertised The annual reports also provide an ‘80/80’ assessment; that is, the fraction of users of each of the monitored ISPs who receive more than 80% of the advertised speed during at least 80% of the peak period (see Figure 5). 74 US FCC (2016), online data at https://www.fcc.gov/reports-research/reports/measuring-broadband-america/charts- measuring-broadband-america-2016#chart5 (viewed 10 May 2017). This corresponds to Chart 5 in the report. Ref: 2009152-254 .
Study on net-neutrality regulation | 31 Figure 5: The ratio of ‘80/80’ consistent median download speed to advertised download speed [Source: Measuring Broadband America programme online data, 2016 75] 250% 80/80 Download Speed/ Advertised Actual Download Speed/ Download Speed (%) Advertised Download Speed (%) 200% 150% 100% 50% 0% Verizon Fiber Comcast TWC Hughes Windstream Frontier Fiber CenturyLink Charter AT&T - IPBB Verizon DSL Optimum Frontier DSL Cox AT&T - DSL Mediacom ViaSat/Exede DSL Cable Fiber Satellite The FCC’s reports provide additional analysis of parameters such as packet loss and latency (see Figure 6). The latency data distinguish between terrestrial ISPs and satellite ISPs, since the round- trip time to a geosynchronous satellite is hundreds of milliseconds (due to the propagation delay for light in a vacuum). Figure 6: Latency by ISP (terrestrial ISPs), 2016 [Source: Measuring Broadband America programme online data, 201675] 60 50 Latency in miliseconds 40 30 20 10 0 AT&T - DSL AT&T - IPBB CenturyLink Frontier DSL Verizon DSL Windstream Optimum Charter Comcast Cox Mediacom TWC Frontier Verizon Fiber Fiber DSL Cable Fiber 75 US FCC (2016), online data available at http://data.fcc.gov/download/measuring-broadband-america/2016/chart6- fixed-2016.xlsx (viewed 10 May 2017). This corresponds to Chart 6 in the report. Ref: 2009152-254 .
Study on net-neutrality regulation | 32 Under the Open Internet Order of 2015, ISPs are required to disclose speed, latency and packet loss. In relation to speed, the Order refers to two distinct measures: “both expected and actual download 76 and upload speeds, latency, and packet loss for each service”. Network operators that voluntarily participate in the Measuring Broadband America programme and make their results public are deemed to have complied with these reporting requirements, and thus granted “safe harbour”. 76 Paras. 165–166. See FCC (2016), Guidance on Open Internet Transparency Rule Requirements, which goes on to note that the Measuring Broadband America programme “measures speed by the throughput over a five second time window, latency by the round trip time between an end user and an off-net measurement server, and packet loss by the percentage of packets transmitted from an end user to a measurement server for which no acknowledgement was received.” This FCC document provides moderately detailed guidance on what should be reported. Ref: 2009152-254 .
Study on net-neutrality regulation | 33 5 Lessons learnt and concluding remarks The analysis undertaken for this study provides the following general lessons which may be of use for European NRAs and BEREC while they are implementing EU net-neutrality regulation: • Quality of service (QoS) clearly has an impact on net neutrality, but the two concepts are quite distinct. Regulators in some markets are actively monitoring QoS, but have chosen to rely on a range of different methods to monitor net neutrality, such as qualitative reporting of traffic- management practices by the ISPs, the ‘seeding’ of third-party initiatives, and having an established complaint system. • The important role of the complaint systems in the USA and Chile means that enforcement in these two countries has a primarily ex-post character (in contrast to the EU, where Regulation 2015/2120 requires NRAs to monitor proactively, on an ex-ante basis). • In some regulatory regimes, existing mechanisms and powers for tackling anti-competitive behaviour are chosen to be relied upon for some non-net-neutral practices, which provides regulators with future discretion. In Chile, tolerance of traffic management is checked against competition principles. In the USA, the FCC relies on telecoms law that was designed prior to the Internet era, to ensure reasonableness of prices, prohibit discrimination, respond to complaints, and impose penalties • Third-party organisations can provide useful complements to the NRAs, in terms of expertise and capacity building in measurement systems suitable for the detection of certain types of net- neutrality violations. Examples include the University of Chile and the Broadband Internet Technical Advisory Group (BITAG) in the USA. • Regulators are considering non-net neutrality practices across both fixed and mobile networks. In the case of mobile networks, crowd-sourced smartphone app-based approaches for gathering measurements are being used in both Chile and the USA. • Many different tools are available to detect practices which may violate net neutrality (either ex ante or ex post), but it is unlikely that any single tool can provide a comprehensive solution. Incidents such as the Comcast/BitTorrent episode in the USA 77 (see Section 3.2.3) demonstrate that regulators will have difficulty anticipating every possible form of net-neutrality abuse in advance. In the EU/EEA, NRAs are obliged to actively monitor non-net-neutral practices themselves. The technical requirements of doing so need to be carefully thought through and specified, and multiple tools or methods are likely to be required. This may imply the need, not for a tool, but for a toolkit that can grow over time as new risks are identified and as new forms of abuse are encountered. 77 Comcast had been blocking peer-to-peer network uploads, a form of degradation that had not been anticipated by regulatory experts. It was fortuitous that Rob Topolski, a network engineer, amateur musician and broadband subscriber happened to have a packet monitor and the competence to understand what he was seeing. Ref: 2009152-254 .
Study on net-neutrality regulation | A–1 Annex A Tools and techniques available to detect and characterise non-net-neutral practices In this annex we provide a wide-ranging review of tools and techniques available to detect and characterise non-net-neutral practices. A wide range of tools and techniques were of interest to the study, including tools and techniques: 1. deployed or under deployment by NRAs 2. imposed or suggested by legal or regulatory frameworks, but not yet deployed or under deployment 3. made publicly available by third parties but currently not imposed, suggested, deployed or under deployment 4. that could be thought of (concepts), but are not currently made publicly available by third parties nor imposed, suggested, deployed or under deployment. Because many measurement tools and concepts are publicly available on the global Internet, categories 3 and 4 above required consideration of any publicly available measurement tools relevant to practices and mechanisms which may violate net neutrality (not just those considered in the benchmark countries). In Figure 7 overleaf we provide a list of publicly available measurement tools (and concepts), including a high-level summary of each. Ref: 2009152-254 .
Study on net-neutrality regulation | A–2 Figure 7: List of tools and techniques for detecting and characterising non-net-neutral practices [Source: Analysys Mason, 2017] Name Category Description Purpose and capabilities Netalyzr Software Netalyzr is a Java applet that runs on the end user’s device and Netalyzr runs a series of tests that collectively take a tool communicates with ICSI servers to measure performance through the few minutes. The tests verify a wide range of properties network. of the end-user machine and the network to which it is connected. See Section 4.3. NDT 78 Software NDT (Network Diagnostic Tool) provides network configuration and NDT is designed to identify a set of performance tool performance testing to Internet users. The system uses the problems, by analysing the specific conditions communication between the client program and a server to perform associated with them (such as duplex mismatch diagnostic functions, through a set of tests, about the poor performance conditions on Ethernet links or incorrectly set TCP of the network. As well as reporting upload and download speeds, NDT buffers in the user’s computer). also tries to understand the problems that caused limited speeds. Glasnost 79 Software Glasnost is a measurement tool that attempts to detect whether users Glasnost can detect traffic shaping in both upstream and tool are subject to traffic differentiation. Its main objective is to make ISPs’ downstream directions separately. traffic-shaping policies transparent to their users. It can be used to test Glasnost’s tests focus on identifying whether an whether an ISP is throttling or blocking email, SSH, Flash Video, HTTP application is being throttled or blocked. or P2P applications such as eMule. Tests can also identify whether application flows are shaped in terms of their port numbers or their packets’ payload. ShaperProbe 80 Software ShaperProbe allows users to detect whether their ISP is applying a ShaperProbe aims to identify whether ISPs are tool token-bucket shaping method to their traffic, and measure the extent of classifying certain kinds of traffic as “low priority”, and shaping. (A token bucket allows a certain number of bytes to be provided offering a different QoS for those flows of traffic. at the peak capacity of the link, whilst any remaining traffic is serviced at ShaperProbe can detect the shaping rate and the a lower rate.) maximum burst before shaping begins. ShaperProbe begins by measuring the link’s capacity, and then it tries to detect if traffic is being shaped on the link in both upstream and downstream directions. 78 See http://software.internet2.edu/ndt/ 79 See http://broadband.mpi-sws.org/transparency/bttest-mlab.php 80 See http://netinfer.net/diffprobe/shaperprobe.html Ref: 2009152-254
Study on net-neutrality regulation | A–3 Name Category Description Purpose and capabilities Neubot 81 Software Neubot (the network neutrality bot) is free software for gathering useful Although Neubot does not detect net-neutrality tool information to study net neutrality. Neubot performs periodic network- violations by itself, all information gathered is released performance measurements. Three types of types of test are currently into the public domain and is available to be analysed by implemented: any person, allowing the investigation of net-neutrality • HTTP and BitTorrent performance issues. • raw TCP performance • MPEG DASH streaming emulation. The tests measure the application-level throughput and round-trip time (RTT) in both downlink and uplink directions. The architecture is configurable, so the user can select what tests should be carried out, at what time and for how long. NANO 82 Software NANO (Network Access Neutrality Observatory) uses a combination of NANO’s installed agents examine the number of tool agents installed at participating clients across the Internet and statistical packets per second sent for each active flow, as well as analysis to identify whether an ISP is discriminating against a specific looking for unexpected problems such as packet loss or service or group of clients. NANO aggregates passive measurements TCP reset packets. They also monitor the load on the from multiple end users and stratifies the measurements to evaluate the client computer. impact of multiple factors. The main objective of NANO is to identify relationships between performance degradations and an ISP’s policy. RIPE Atlas 83 Hardware RIPE Atlas is a global, open, distributed Internet measurement network RIPE Atlas continuously monitors the reachability of a based tool that provides a real-time perception of the state of the Internet. RIPE network or host, via the thousands of HW devices Atlas collects and releases into the public domain connectivity and deployed around the world. reachability data from thousands of devices all over the world. The It allows the investigation and troubleshooting of hardware devices collect data on the Internet access of voluntary users. reported network problems, by conducting ad-hoc These devices are USB-powered and connected to the host’s router, connectivity checks. allowing them to perform active measurements including ping, traceroute, DNS, SSL/TLS and NTP. 81 See http://www.neubot.org/ 82 See https://noise-lab.net/projects/old-projects/nano/ 83 See https://atlas.ripe.net/ Ref: 2009152-254
Study on net-neutrality regulation | A–4 Name Category Description Purpose and capabilities M-Lab 84 Hardware M-Lab is a consortium of research, industry and public-interest partners M-Lab is not a measurement tool, but rather it provides platform dedicated to: 85 the infrastructure to set up a measurement platform on • “Providing an open, verifiable measurement platform for global which measurement tools can be deployed. network performance” All the data collected by those tools is released into the • “Hosting the largest open Internet performance dataset on the planet” public domain. • “Creating visualizations and tools to help people make sense of Internet performance”. Some of the tools available on the M-Lab platform are NDT, Neubot, MobilePerf and Glasnost. NetPolice 86 Software NetPolice enables the detection of content- and routeing-based NetPolice detects traffic management that induces tool differentiations in backbone ISPs. This tool is mainly used by large users, packet loss. It performs loss-rate measurements to such as access providers or content providers, rather than by end users. detect traffic differentiation. NPAD 87 Software NPAD (Network Path & Application Diagnostics) is a tool to diagnose end NPAD runs simple TCP tests, including throughput, RTT tool users’ network-performance issues affecting their device or the network or packet-loss metrics. between the end user and the nearest NPAD server. For each diagnosed problem, the server prescribes corrective actions with instructions that are easy for non- experts to understand. BISmark 88 Software BISmark (Broadband Internet Service Benchmark) is an open platform BISmark software performs several active tool which aims to measure the QoS performance delivered to ISPs’ end measurements such as latency, packet loss, jitter, and users, visualising and monitoring traffic patterns through the devices downlink and uplink throughput. inside their home network. The BISmark application is available for: It provides a dashboard showing performance over time • Android phones and comparing performance across ISPs and regions. • Raspberry PI All data collected by BISmark is released into the public • OpenWrt routers. domain. 84 See https://www.measurementlab.net/ 85 See https://www.measurementlab.net/about/ 86 See https://web.eecs.umich.edu/~zmao/Papers/netpolice.pdf 87 See http://www.ucar.edu/npad/ 88 See http://www.projectbismark.net/ Ref: 2009152-254
