Comparing DNS Resolvers in the Wild Bernhard Ager Wolfgang M¨ uhlbauer Georgios Smaragdakis Steve Uhlig { bernhard|steve|georgios } @net.t-labs.tu-berlin.de { wolfgang.muehlbauer@tik.ee.ethz.ch } Technische Universtit¨ at Berlin/Deutsche Telekom Laboratories ETH Z¨ urich November 1 2010 Ager et al (TUB/DT Labs/ETH) Comparing DNS Resolvers in the Wild Nov 1 2010 1
Motivation Domain Name System (DNS) DNS: resolve www.fg-inet.de to 85.214.78.63 Send HTTP request through Internet towards 85.214.78.63 ⇒ Fundamental building block of the Internet ⇒ Its performance is critical Ager et al (TUB/DT Labs/ETH) Comparing DNS Resolvers in the Wild Nov 1 2010 2
Motivation Motivation DNS according to the text book A commodity service: everyone is using it Resolves hostnames to IP addresses Same view from everywhere Caching Resolver assigned by ISP Ager et al (TUB/DT Labs/ETH) Comparing DNS Resolvers in the Wild Nov 1 2010 3
Motivation Motivation DNS according to the text book DNS in the wild A commodity service: New (mis)uses of DNS [Vixie’09] everyone is using it Locality aware replies Dynamic load balancing Resolves hostnames to IP NXDOMAIN catching addresses Use as directory service Same view from everywhere Third party resolvers, e. g., Caching Google Public DNS and Resolver assigned by ISP OpenDNS [Vixie’09] P. Vixie. What DNS is not. Commun. ACM , 52(12):43–47, 2009. Ager et al (TUB/DT Labs/ETH) Comparing DNS Resolvers in the Wild Nov 1 2010 3
Motivation Motivation DNS according to the text book DNS in the wild A commodity service: New (mis)uses of DNS [Vixie’09] everyone is using it Locality aware replies Dynamic load balancing Resolves hostnames to IP NXDOMAIN catching addresses Use as directory service Same view from everywhere Third party resolvers, e. g., Caching Google Public DNS and Resolver assigned by ISP OpenDNS DNS is changing ⇒ Study across content, locations and resolvers Compare DNS deployment of different ISPs and different resolvers Metrics: responsiveness and quality of replies [Vixie’09] P. Vixie. What DNS is not. Commun. ACM , 52(12):43–47, 2009. Ager et al (TUB/DT Labs/ETH) Comparing DNS Resolvers in the Wild Nov 1 2010 3
Data and Approach Data and Approach Overview Custom script 3 DNS resolvers: Google DNS, OpenDNS, local resolver 10k+ hostnames: Popular content: Top 5000 from Alexa Less popular content: Bottom 2000 from Alexa Many objects on websites: 3000+ “embedded” hostnames Two back-to-back queries for each hostname for each resolver ⇒ > 60 traces from all around the globe, > 50 different ISPs Ager et al (TUB/DT Labs/ETH) Comparing DNS Resolvers in the Wild Nov 1 2010 4
Results First vs. second query time The expected 5000 GOOGLE Scatterplot LOCAL 2000 OPENDNS 5000 ms timeout Time for second query (ms) Minima indicate RTT to 500 DNS server 200 Small variance for second 100 query due to caching 50 ⇒ Local DNS apparently better 20 than OpenDNS and GoogleDNS 10 10 20 50 100 500 2000 5000 Time for first query (ms) Ager et al (TUB/DT Labs/ETH) Comparing DNS Resolvers in the Wild Nov 1 2010 5
Results First vs. second query time The expected and the unexpected 5000 5000 GOOGLE GOOGLE LOCAL LOCAL 2000 OPENDNS OPENDNS 2000 Time for second query (ms) Time for second query (ms) 500 500 200 100 100 50 50 20 20 10 10 10 20 50 100 500 2000 5000 10 20 50 100 500 2000 5000 Time for first query (ms) Time for first query (ms) Ager et al (TUB/DT Labs/ETH) Comparing DNS Resolvers in the Wild Nov 1 2010 5
Results Impact of redirection How many replies are in the same AS as the vantage point? Google Local 800 OpenDNS # returned IPs that are in same AS Distant DNS servers 600 CDN optimizes for resolver IP address 400 Up to 1000 hosts with local resolval available 200 not locally resolved with GoogleDNS/OpenDNS 0 vantage points (sorted by # returned IPs that are in same AS) Ager et al (TUB/DT Labs/ETH) Comparing DNS Resolvers in the Wild Nov 1 2010 6
Summary Summary Summary Application performance depends on DNS deployment Local DNS not always the performance champion But for content locality you have to use local DNS DNS is a “battlefield” with many players How do we get to an open DNS system that can cope with today’s needs? Ager et al (TUB/DT Labs/ETH) Comparing DNS Resolvers in the Wild Nov 1 2010 7
Summary Current work Current work Repeat experiment in wider scope ⇒ Please help and run our new measurement script from http://www.fg-inet.de/ at your home No privacy implications No interference with your personal data Source code fully open Manual trace upload Little overhead Ager et al (TUB/DT Labs/ETH) Comparing DNS Resolvers in the Wild Nov 1 2010 8
Summary Current work Current work Repeat experiment in wider scope ⇒ Please help and run our new measurement script from http://www.fg-inet.de/ at your home No privacy implications No interference with your personal data Source code fully open Manual trace upload Little overhead Questions? Ager et al (TUB/DT Labs/ETH) Comparing DNS Resolvers in the Wild Nov 1 2010 8
Recommend
More recommend
