Deferred Peerings as Obstacles on the Internet Thomas Rieder - PowerPoint PPT Presentation

Deferred Peerings as Obstacles on the Internet Thomas Rieder HTTPS://PEERING.RIEDER.IO

Discussion Motivation Fundamentals Design Methodology Conclusion Motivation Pe Peering : Direct connection between two networks on the Internet (“interconnection”) → Defines the topology of the Internet → Goal is better connection quality or redundancy Ne Net neut neutrality : provider deliberately slows down traffic Peering : provider decides against speeding it up Pe HTTPS://PEERING.RIEDER.IO

Discussion Motivation Fundamentals Design Methodology Conclusion Motivation Peering is settlement-free Minimal setup effort Basically no technical / economical reason not to do it → Some providers still decide against it Go Goal al : identify those providers and their reasons HTTPS://PEERING.RIEDER.IO

Discussion Motivation Fundamentals Design Methodology Conclusion Fundamentals Autonomous Systems Structure of the Internet Different network tiers Internet Exchange Points BGP RIPE Atlas HTTPS://PEERING.RIEDER.IO

Discussion Motivation Fundamentals Design Methodology Conclusion Autonomous System “Network” on the Internet Entity that routing is based on Uniquely identified by A utonomous S ystem N umber (ASN) Handed out by R egional I nternet R egistries (RIR) Consists of multiple IP prefixes 192.189.5 1 .0/24 192.189.5 3 .0/24 HTTPS://PEERING.RIEDER.IO

Discussion Motivation Fundamentals Design Methodology Conclusion Autonomous System “Network” on the Internet Entity that routing is based on Uniquely identified by A utonomous S ystem N umber (ASN) Handed out by R egional I nternet R egistries (RIR) Consists of multiple IP prefixes Mapping IP ↔ ASN via WHOIS or RDAP HTTPS://PEERING.RIEDER.IO

Discussion Motivation Fundamentals Design Methodology Conclusion Structure of the Internet HTTPS://PEERING.RIEDER.IO

Discussion Motivation Fundamentals Design Methodology Conclusion Structure of the Internet HTTPS://PEERING.RIEDER.IO

Discussion Motivation Fundamentals Design Methodology Conclusion Structure of the Internet HTTPS://PEERING.RIEDER.IO

Discussion Motivation Fundamentals Design Methodology Conclusion Structure of the Internet - IXP HTTPS://PEERING.RIEDER.IO

Discussion Motivation Fundamentals Design Methodology Conclusion Structure of the Internet HTTPS://PEERING.RIEDER.IO

Discussion Motivation Fundamentals Design Methodology Conclusion Structure of the Internet HTTPS://PEERING.RIEDER.IO

Discussion Motivation Fundamentals Design Methodology Conclusion B order G ateway P rotocol (BGP) Routing protocol of the Internet Determines how to get packets to the target network Operates on paths → computes single-source shortest path to all other AS No Non-publ public routing policies for political / economical / operational constraints HTTPS://PEERING.RIEDER.IO

Discussion Motivation Fundamentals Design Methodology Conclusion RIPE Atlas We want to analyze multiple ISPs It’s unfeasible to lease a connection from every single one → RI RIPE Atl tlas : global Internet measurement network with 9,000 probes → Allows us to do traceroute measurements from a lot of different sources HTTPS://PEERING.RIEDER.IO

Discussion Motivation Fundamentals Design Methodology Conclusion RIPE Atlas - Probes HTTPS://PEERING.RIEDER.IO

Discussion Motivation Fundamentals Design Methodology Conclusion HTTPS://PEERING.RIEDER.IO

Discussion Motivation Fundamentals Design Methodology Conclusion Available at https://peering.rieder.io HTTPS://PEERING.RIEDER.IO

Discussion Motivation Fundamentals Design Methodology Conclusion Methodology Focus on a single country at a time Execute traceroutes from the RIPE Atlas probes Checks the resulting paths for hops that indicate peering Ta Target : Content Servers (ISP Speed Index) HTTPS://PEERING.RIEDER.IO

Discussion Motivation Fundamentals Design Methodology Conclusion Methodology – Peering Graphs Construct graphs among all networks in a single country Good for analyzing the general approach to peering → traceroutes among all probes HTTPS://PEERING.RIEDER.IO

Discussion Motivation Fundamentals Design Methodology Conclusion Discussion: Austria Only one IXP: Vienna Internet Exchange Not all networks have probes 32 out of 65 networks peer understandable for company networks → We focus on the large consumer networks HTTPS://PEERING.RIEDER.IO

Discussion Motivation Fundamentals Design Methodology Conclusion Austria – Peering with Ap April May Ma June Ju July Ju Au August A1 3.42 3.36 3.45 3.47 3.52 CableLink 3.92 3.93 3.86 3.72 3.82 Kabelplus 4.15 4.13 4.15 4.18 4.24 LIWEST 4.18 4.12 4.13 4.21 4.28 Tele2 2.99 2.96 2.92 2.96 3.01 UPC 2.70 3.07 3.75 3.94 3.98 Al All va values ar are th the co connection sp speed in in Mb Mbps Peering No Peering No Data HTTPS://PEERING.RIEDER.IO

Discussion Motivation Fundamentals Design Methodology Conclusion Austria – UPC HTTPS://PEERING.RIEDER.IO

Origin Network Transit Network Client Network HTTPS://PEERING.RIEDER.IO

Discussion Motivation Fundamentals Design Methodology Conclusion Austria – A1 Only major ISP that refuses to peer No aversion to peering in general They are present at the Vienna Internet Exchange Run Video-On-Demand service themselves HTTPS://PEERING.RIEDER.IO

Discussion Motivation Fundamentals Design Methodology Conclusion Discussion: Italy Two IXPs: Milan and Turin Internet quality is generally not as good as in Austria Netflix is rather new in Italy (October 2015) 33 out of 65 networks peer Layer 2 peering likely HTTPS://PEERING.RIEDER.IO

Origin Network Transit Network Client Network HTTPS://PEERING.RIEDER.IO

Discussion Motivation Fundamentals Design Methodology Conclusion Discussion: The Netherlands By far the best Internet quality Some of the biggest IXP in the world in Amsterdam (by traffic) → Gateway between North America and Europe Only 19 out of 64 networks peer Again: layer 2 peering likely (extremely low latencies) HTTPS://PEERING.RIEDER.IO

Discussion Motivation Fundamentals Design Methodology Conclusion The Netherlands – Peering with April Ap May Ma June Ju July Ju Au August Caiway 4.07 4.03 3.98 4.02 3.98 Ziggo 3.95 3.9 3.87 3.92 3.90 UPC 3.91 3.86 3.85 3.92 3.89 ZeelandNet 3.91 3.93 3.89 3.85 3.81 KPN 3.8 3.8 3.74 3.84 3.82 Xs4all 3.78 3.75 3.73 3.77 3.75 Online 3.48 3.4 3.35 3.41 3.39 Tele2 3.43 3.35 3.36 3.4 3.42 Al All va values ar are th the co connection sp speed in in Mb Mbps Peering No Peering No Data HTTPS://PEERING.RIEDER.IO

Discussion Motivation Fundamentals Design Methodology Conclusion Conclusion Peering can have a big impact on performance (up to 47%) RIPE Atlas allows us to measure it Caveat: Layer 2 peering can only be inferred from the latency No reason not to peer Some companies still refuse to peer in order to promote their own services Opposed to net neutrality it’s more a sin of omission rather than direct manipulation HTTPS://PEERING.RIEDER.IO

Thank You! Thomas Rieder HTTPS://PEERING.RIEDER.IO

Discussion Motivation Fundamentals Design Methodology Conclusion Motivation Peering affects how fast content can be delivered to users Similar to net neutrality, but works differently Ne Net neutrality : provider deliberately slows down traffic Pe Peering : provider decides against speeding it up HTTPS://PEERING.RIEDER.IO

Discussion Motivation Fundamentals Design Methodology Conclusion Structure of the Internet Pr Private Interco connect HTTPS://PEERING.RIEDER.IO

Discussion Motivation Fundamentals Design Methodology Conclusion RIPE Atlas - Coverage HTTPS://PEERING.RIEDER.IO

Discussion Motivation Fundamentals Design Methodology Conclusion RIPE Atlas - Features Allows us to do traceroute measurements from a lot of different sources Based on a credit-system We need to earn credits first Limits how many measurements we can do HTTPS://PEERING.RIEDER.IO

HTTPS://PEERING.RIEDER.IO

Origin Network Transit Network Client Network HTTPS://PEERING.RIEDER.IO

Discussion Motivation Fundamentals Design Methodology Conclusion Structure of the Internet – Route Servers HTTPS://PEERING.RIEDER.IO

Discussion Motivation Fundamentals Design Methodology Conclusion Structure of the Internet – Route Servers HTTPS://PEERING.RIEDER.IO

Agenda Motivation Fundamentals Design Methodology Discussion Conclusion HTTPS://PEERING.RIEDER.IO

Discussion Motivation Fundamentals Design Methodology Conclusion Italy – Peering with April Ap May Ma June Ju July Ju Au August EOLO – NGI 2.87 2.85 2.9 2.88 2.97 Fastweb 3.34 3.3 3.31 3.42 3.44 Linkem 2.13 2.17 2.26 2.46 2.49 Telecom Italia 3.17 3.15 3.09 3.11 3.13 Tiscali 3.16 3.07 3.12 3.18 3.18 Vodafone 3.26 3.29 3.28 3.35 3.36 Vodafone TeleTu 2.67 2.58 2.59 2.61 2.59 Wind 3.14 3.12 3.1 3.11 3.12 Al All va values ar are th the co connection sp speed in in Mb Mbps Peering No Peering No Data HTTPS://PEERING.RIEDER.IO