Sunday, February 3, 2013

Week4:DNS Performance and the Effectiveness of Caching

----For EECS345 Distributed System (CS Northwestern course) weekly reading report

                            Week4:DNS Performance and the Effectiveness of Caching

       This paper has presented a detailed analysis of traces of DNS and associated TCP traffic collected on the Internet based on the data form traced links of MIT Laboratory and KAIST, then discusses trace-driven simulations to study the effectiveness of DNS caching as a function of TTL and degree of cache sharing.

       By analyse the collected data, the authors have these results:
1.                 1.  Distribution of popular names following the Zipf’s law fails to make use of caching with larger TTL values.
2.                 2.  Sharing cache among groups of clients has limited gain in terms of cache hit after the total member count crosses 20-25.
3.                3. The client-perceived latency is adversely affected by number of referrals, and caching NS records to reduce the number of referrals will decrease latency as well as load on the root servers.
4.               4.  Distribution of names causing negative responses follows a heavy tailed distribution as well. As a result, hit rate of negative caching is also limited.

       Using trace-driven simulations algorithm, the author wants to find how useful to share DNS caches among many client machines and what is the impact of choice of TTL on caching effectiveness. The authors quantify two important statistics:  the distribution of name popularity and TTL values in the trace data. Both determine cache hit rates. And draw a conclusion that for A-records, lower TTL seems won’t harm the hit rates, caching appears to have limited effectiveness. But for NS-records, it would increase the load on root server and harm DNS scalability.

At the end, the paper draw a conclusion that the  widespread use of dynamic, lower-TTL A-record bindings should not harm DNS performance. Such that the scalability of DNS are less dependent on the hierarchical design of its name space or good A-record caching(originally believed).

No comments:

Post a Comment