John S. Otto Mario A. Sánchez John P. Rula Fabián E. Bustamante

Similar documents
Mario A. Sánchez John S. Otto Zachary S. Bischof Fabián E. Bustamante

Fabián E. Bustamante

Content Distribution. Today. l Challenges of content delivery l Content distribution networks l CDN through an example

Factors Affecting Performance of Web Flows in Cellular Networks

CSE 124: CONTENT-DISTRIBUTION NETWORKS. George Porter December 4, 2017

CONTENT-DISTRIBUTION NETWORKS

Cell Spotting: Studying the Role of Cellular Networks in the Internet. John Rula. Fabián E. Bustamante Moritz Steiner 2017 AKAMAI FASTER FORWARD TM

ACMS: The Akamai Configuration Management System. A. Sherman, P. H. Lisiecki, A. Berkheimer, and J. Wein

From Internet Data Centers to Data Centers in the Cloud

CONTENT-AWARE DNS. IMPROVING CONTENT-AWARE DNS RESOLUTION WITH AKAMAI DNSi CACHESERVE EQUIVALENCE CLASS. AKAMAI DNSi CACHESERVE

416 Distributed Systems. March 23, 2018 CDNs

A measurement experimentation platform at the Internet s edge

Factors Affecting Performance of Web Flows in Cellular Networks

End-user mapping: Next-Generation Request Routing for Content Delivery

11/13/2018 CACHING, CONTENT-DISTRIBUTION NETWORKS, AND OVERLAY NETWORKS ATTRIBUTION

Democratizing Content Publication with Coral

Characterizing a Meta-CDN

Democratizing Content Publication with Coral

Distributed Systems. 21. Content Delivery Networks (CDN) Paul Krzyzanowski. Rutgers University. Fall 2018

CS November 2018

Networked systems and their users

Drafting Behind Akamai (Travelocity-Based Detouring)

Web as a Distributed System

Internet Inter-Domain Traffic

Dasu: Pushing Experiments to the Internet s Edge

Dasu: Pushing Experiments to the Internet s Edge

Files/News/Software Distribution on Demand. Replicated Internet Sites. Means of Content Distribution. Computer Networks 11/9/2009

CSE 123b Communications Software

Today s class. CSE 123b Communications Software. Telnet. Network File System (NFS) Quick descriptions of some other sample applications

Overview Computer Networking Lecture 16: Delivering Content: Peer to Peer and CDNs Peter Steenkiste

Web Caching and Content Delivery

Fixing the Embarrassing Slowness of OpenDHT on PlanetLab

Venugopal Ramasubramanian Emin Gün Sirer SIGCOMM 04

Trying broadband characterization at home

Internet Content Distribution

CS 43: Computer Networks. 14: DHTs and CDNs October 3, 2018

CS 43: Computer Networks BitTorrent & Content Distribution. Kevin Webb Swarthmore College September 28, 2017

CS November 2017

Opportunities for Exploiting Social Awareness in Overlay Networks. Bruce Maggs Duke University Akamai Technologies

Chapter 6: Distributed Systems: The Web. Fall 2012 Sini Ruohomaa Slides joint work with Jussi Kangasharju et al.

The hidden locality in swarms

TMA workshop 2010 April

Department of Computer Science Institute for System Architecture, Chair for Computer Networks. Caching, Content Distribution and Load Balancing

Network Working Group

A Tale of Three CDNs

CSCE 463/612 Networks and Distributed Processing Spring 2018

Characterization of Collaborative Resolution in Recursive DNS Resolvers

OpenCache. A Platform for Efficient Video Delivery. Matthew Broadbent. 1 st Year PhD Student

taming the Torrent Over the past decade, the peer-topeer Fabián E. Bustamante ;LOGIN: February 2010 Taming the Torrent 51

TAIL LATENCY AND PERFORMANCE AT SCALE

CSC2231: DNS with DHTs

Naming and Content Distribution

Mul$media Networking. #9 CDN Solu$ons Semester Ganjil 2012 PTIIK Universitas Brawijaya

Exam - Final. CSCI 1680 Computer Networks Fonseca. Closed Book. Maximum points: 100 NAME: 1. TCP Congestion Control [15 pts]

SaaS Providers. ThousandEyes for. Summary

0 0& Basic Background. Now let s get into how things really work!

Performance Characterization of a Commercial Video Streaming Service. Mojgan Ghasemi, Akamai Technologies - Princeton University

Domain Name System.

Mobile Content Hosting Infrastructure in China: A View from a Cellular ISP. Zhenhua Li Chunjing Han Gaogang Xie

Odin: Microsoft s Scalable Fault-

The Guide to Best Practices in PREMIUM ONLINE VIDEO STREAMING

A content delivery perspective on mobility in the Internet

1752 IEEE/ACM TRANSACTIONS ON NETWORKING, VOL. 17, NO. 6, DECEMBER 2009

Mapping the Expansion of Google s Serving Infrastructure

Understanding and Characterizing Hidden Interception of the DNS Resolution Path

Distributed Systems Principles and Paradigms. Chapter 12: Distributed Web-Based Systems

Distributed Systems Architectures

List of measurements in rural area

CSE 124: QUANTIFYING PERFORMANCE AT SCALE AND COURSE REVIEW. George Porter December 6, 2017

A Survey on Research on the Application-Layer Traffic Optimization (ALTO) Problem

working with Akamai to improve HTTP(S) traffic flows

CSE 473 Introduction to Computer Networks. Exam 1. Your name here: 10/3/2012

ThousandEyes for. Application Delivery White Paper

CSE 124: TAIL LATENCY AND PERFORMANCE AT SCALE. George Porter November 27, 2017

Follow me!

Server Selection Mechanism. Server Selection Policy. Content Distribution Network. Content Distribution Networks. Proactive content replication

Application Layer: OSI and TCP/IP Models

Data Center Performance

CSc 450/550 Computer Networks Domain Name System

Multimedia Streaming. Mike Zink

Characterization of Collaborative Resolution in Recursive DNS Resolvers

PNUTS: Yahoo! s Hosted Data Serving Platform. Reading Review by: Alex Degtiar (adegtiar) /30/2013

Defining the Impact of the Edge Using The SamKnows Internet Measurement Platform. PREPARED FOR EdgeConneX

Where s my DNS? Sara Dickinson IDS 2. Where s my DNS?

Reverse Traceroute. NSDI, April 2010 This work partially supported by Cisco, Google, NSF

IPv6 at Google. Lorenzo Colitti

THE UTILITY OF DNS TRAFFIC MANAGEMENT

Networking Named Content: Content-Centric Networking. John Rula

SamKnows test methodology

Drongo. Speeding Up CDNs with Subnet Assimilation from the Client. CoNEXT 17 Incheon, South Korea CDN & Caching Session

AWS Well Architected Framework

On the Use and Performance of Content Distribution Networks

Electrical Engineering and Computer Science Department

ECE 650 Systems Programming & Engineering. Spring 2018

Telematics Chapter 9: Peer-to-Peer Networks

EECS 122: Introduction to Computer Networks DNS and WWW. Internet Names & Addresses

Deploying the BIG-IP System v10 with Oracle s BEA WebLogic

Complex Interactions in Content Distribution Ecosystem and QoE

Document Sub Title. Yotpo. Technical Overview 07/18/ Yotpo

Joint Server Selection and Routing for Geo-Replicated Services

Transcription:

John S. Otto Mario A. Sánchez John P. Rula Fabián E. Bustamante Northwestern, EECS http://aqualab.cs.northwestern.edu

! DNS designed to map names to addresses Evolved into a large-scale distributed system! CDNs leverage DNS for dynamic routing Assume proximity between users and their resolvers! Use of remote DNS Servers concentrated farther from users Susceptible to configuration errors (e.g. Comcast DNS outages)! Growing alternative third-party DNS services Public DNS usage has grown to 11% of users! So what? IMC 12 Otto et al. Content Delivery and the Natural Evolution of DNS! 2

Visit cnn.com 34 DNS lookups 204 HTTP requests 520 KB of data downloaded IMC 12 Otto et al. Content Delivery and the Natural Evolution of DNS! 3

56% of domains resolve to a CDN IMC 12 Otto et al. Content Delivery and the Natural Evolution of DNS! 4

! It s not just cnn.com % using CDNs 100 90 80 70 60 50 40 30 20 10 0 74% of the top 1000 web sites use CDNs 10 50 100 500 1000 N most popular sites Sites Pageviews IMC 12 Otto et al. Content Delivery and the Natural Evolution of DNS! 5

CDN Replica Content Origin Local DNS End Host CDN Replica IMC 12 Otto et al. Content Delivery and the Natural Evolution of DNS! 6

We see 27% annual increase in their usage CDN Replica Content Origin Local DNS End Host CDN Replica Remote DNS Remote DNS services break this assumption IMC 12 Otto et al. Content Delivery and the Natural Evolution of DNS! 7

! The cost of remote DNS! The industry response! An end-host-based solution IMC 12 Otto et al. Content Delivery and the Natural Evolution of DNS! 8

! An experiment in Dasu a BitTorrent-based platform for network characterization and experimentation Subset of clients: 10,923 hosts, 99 countries, 752 ISPs! Measure DNS servers ISP servers and public DNS services Network and application level probes! Obtain CDN redirections for each DNS service Download small web objects hosted by several CDNs Do iterative resolution for baseline In this talk Google DNS and Akamai CDN as examples DNS Latency End-to-End Latency HTTP Latency Send DNS query Get DNS answ HTTP connect Received 1 st byte of object IMC 12 Otto et al. Content Delivery and the Natural Evolution of DNS! 9

! Multiple redirections show typical set of servers Depends on DNS location, CDN load balancing, network conditions! Compare overlap of servers between locations CDN DNS CDN Replica CDN Replica CDN Replica CDN Replica CDN Replica Local DNS Remote DNS End Host IMC 12 Otto et al. Content Delivery and the Natural Evolution of DNS! 10

! Remote DNS services yield radically different redirections Minimal overlap with those seen from the client 1.0 0.8 0.6 CDF 0.4 Perfect information 0.2 0.0 0.0 0.2 0.4 0.6 0.8 1.0 Overlap (redirection similarity) Redirections via client are consistent IMC 12 Otto et al. Content Delivery and the Natural Evolution of DNS! 11

! Remote DNS services yield radically different redirections Minimal overlap with those seen from the client 1.0 0.8 A good approximation 80% median overlap CDF 0.6 0.4 Perfect information ISP DNS 0.2 0.0 0.0 0.2 0.4 0.6 0.8 1.0 Overlap (redirection similarity) IMC 12 Otto et al. Content Delivery and the Natural Evolution of DNS! 12

! Remote DNS services yield radically different redirections Minimal overlap with those seen from the client 1.0 0.8 90% have no overlap CDF 0.6 0.4 Perfect information ISP DNS Google DNS 0.2 0.0 0.0 0.2 0.4 0.6 0.8 1.0 Overlap (redirection similarity)! Different redirections, but does it affect performance? IMC 12 Otto et al. Content Delivery and the Natural Evolution of DNS! 13

! Different redirections mean different performance 80% have <15% penalty IMC 12 Otto et al. Content Delivery and the Natural Evolution of DNS! 14

! Different redirections mean different performance In median case, 65% penalty Top 20% have 200% penalty IMC 12 Otto et al. Content Delivery and the Natural Evolution of DNS! 15

! The cost of remote DNS Yields different CDN redirections to 90% of users Increasing end-to-end web latency by 65% for median user! The industry response! An end-host-based solution IMC 12 Otto et al. Content Delivery and the Natural Evolution of DNS! 16

! Idea: avoid impact of remote DNS by changing localization approach directly provide client location to CDN! Implemented as an EDNS0 extension edns-client-subnet! DNS resolver adds client s IP prefix to request! CDN redirection based on client s location, not resolver s! First evaluation of EDNS effectiveness! Approximate client location approach typically sufficient /16 prefix enough for Google and EdgeCast CDNs IMC 12 Otto et al. Content Delivery and the Natural Evolution of DNS! 17

! Focus on places where remote DNS affects performance 45% performance improvement IMC 12 Otto et al. Content Delivery and the Natural Evolution of DNS! 18

! Minor issue both DNS and CDN services must support it % of sites 100 90 80 70 60 50 40 30 20 10 0 No support Google service or CDNsupport CDN support Only 1% of sites use it for content 10 50 100 500 1000 N most popular sites most sites don t support it Most support seen for Google services Should we just wait? 5-10 public DNS 100 s of ISPs 50+ CDNs IMC 12 Otto et al. Content Delivery and the Natural Evolution of DNS! 19

! The cost of remote DNS Yields different CDN redirections to 90% of users Increasing end-to-end web latency by 65% for median user! The industry response edns-client-subnet First evaluation of its effectiveness Median user could get 45% performance improvement Nearly 2 years since proposed; 1% of sites support it! An end-host-based solution IMC 12 Otto et al. Content Delivery and the Natural Evolution of DNS! 20

! Different approach: move the resolver close to the user! End host directly queries for CDN redirection CDN redirection based on client s location! Run a DNS proxy on the user s machine! Monitor stream of requests to identify CDN redirections! Use Direct Resolution to improve redirection quality IMC 12 Otto et al. Content Delivery and the Natural Evolution of DNS! 21

! Step 1: typical DNS query to recursive resolver Use recursive DNS to translate customer name to CDN! Step 2: directly query CDN for an improved redirection DR map CDN Replica 2 CDN DNS End Host 1 Remote DNS CDN Replica Initial map IMC 12 Otto et al. Content Delivery and the Natural Evolution of DNS! 22

! Focus on places where remote DNS affects performance Available now works with all CDNs and DNS services Within 16% of potential Improves performance in 76% of locations IMC 12 Otto et al. Content Delivery and the Natural Evolution of DNS! 23

! The cost of remote DNS Gives different CDN redirections to 90% of users Increases end-to-end web latency by 65% for median user! The industry response edns-client-subnet First evaluation of its effectiveness Median user could get 45% performance improvement Nearly 2 years since proposed; 1% of sites support it! An end-host-based solution 40% improvement for median user Gives better performance in 76% of affected locations Readily available IMC 12 Otto et al. Content Delivery and the Natural Evolution of DNS! 24

! more than just better CDN performance Faster lookups with proactive caching Automatic, personalized server selection Graceful handling of DNS outages! First 23 days! 13,800 users! 125 countries! Get it today! http://aqualab.cs.northwestern.edu/projects/namehelp IMC 12 Otto et al. Content Delivery and the Natural Evolution of DNS! 25