Uploaded by giannis.m.giakoumakis

DNS

advertisement
Comparing DNS Resolvers in the
Wild
DNS
• Domain Name System
• Translates domain
names into IP addresses
• Hierarchical distributed
System
• Nodes are called as
name servers
• What are resolvers?
Comparing DNS Resolvers in the Wild
•
•
•
•
DNS has evolved a lot over the years
From a naming service to commodity infrastructure
Many third party DNS resolvers have been developed
E.g. OpenDNS extends DNS features by adding
misspelling correction, phishing protection, and content
filtering
• Do these third party DNS resolvers really improve
performance?
Content Delivery Networks
• Aim is to provide content with higher availability and
performance
• Brings content closer to the user by using carefully
positioned global infrastructure
• This global infrastructure includes thousands of servers
spread world wide
• Users, when request certain content, are directed to
closest CDN server having that data
• Assumes that DNS resolver is close to the client
originating the request
DNS today : Issues
• CDN’s assumption that DNS resolver is close to the client
originating the request may degrade performance
• A third party DNS which is not in network on the client
may direct it to a CDN server out of it’s network
• CDN’s associate a short TTL on their DNS answers making
caching useless
• “NXDOMAIN catcher” can also hamper some
applications like email
DNS today : Issues -> NXDOMAIN catcher
• NXDOMAIN -> NonExistent Domain
• NXDOMAIN status code
is caught and the IP
address of a certain
search website is
returned
• Advertisements can be
inserted to get profit
Measurements
• We need to compare responsiveness of various DNS
resolvers
• Local Resolver
• OpenDNS
• Google
• Done by performing DNS queries to 10000 hosts.
• Following information collected during the measurements
• Vantage Point
• Resolver
• Host
Host Selection
•
•
•
•
DNS relies on caching for efficiency
Selection done by mix of best and worst sites.
Download top 1,000,000 list of sites from Alexa
10,000 hosts selected as follows:
• top5000
Most popular hosts which are likely to be in cache
• tail2000
Less likely in the cache
• embedded
Web-pages containing flash content that browser may have
to retrieve from different domain
Inside facts of the measurements…
• Two hosts in same area can query DNS back to
back
• Response time will be shorter for the second
host as the reply will be in cache
• Compares response times between first and
second query
• Need to inspect timestamps in traces to find
location
• Traces need to be selected that do not interact
Evaluation of DNS resolvers
• Generally, local DNS resolvers are used
• Third party resolvers claim to provide advanced
services that provide web site and phishing site
blocking capability along with “suggestions” for
failed lookups
• Main problem in third party resolvers – larger
response time
• DNS performance depends of proximity of end hosts
• Caching improves the latency with over 95% of
second queries answered in less than 100ms
Comparison between good ISP and bad ISP
• As can be seen in above diagram, second query has faster
response time due to caching
• For second query in case of bad ISP, local DNS resolver has
almost similar delay as others
DNS Deployment
• “good ISP”: response times for the second queries show only small variation and
are consistently better than those for the first query.
• “bad ISP”: scattered along a horizontal and vertical line, as well as the diagonal. We
explain this behavior by a load balancing setup without a shared cache.
DNS Deployment
• A similar behavior for both OpenDNS and GoogleDNS in
several traces.
• This shows use of load balancing for highly loaded site
• High RTTs towards the local DNS and load balancing:
• DNS infrastructure is centralized and requires load
balancing to compensate for the high number of queries
arriving at a single Location.
• Load balancing is Good,
• But the way some ISPs are implementing it prevents
caching from being properly utilized.
DNS Answers
• Graph shows that local resolver is more likely to return IP
addresses that are in client’s AS than Google or OpenDNS.
• This locally available content covers akamaized set completely
that is local DNS resolvers works better for CDN content.
DNS Answers
• Graph shows number of different IP addresses returned by
Local DNS and Google DNS resolver.
• As, Local resolver returns IP address in the same AS as client,
it can be deduced that Google DNS directs client
unnecessarily out of its network.
Summary
• End-host experiences a very small latency to the
resolvers maintained by the local ISP
• There does exist cases where GoogleDNS and
OpenDNS outperform
• Several ISPs and OpenDNS rely on a load balancing
setup without a shared cache, resulting in poor
caching efficiency.
• Third-party DNS resolvers do not manage to redirect
the users towards content available within the ISP,
contrary to the local DNS ones.
Just The Beginning…
Thank You
Download