J. Jung, E. Sit, H. Balakrishnan, "DNS Performance and the Effectiveness of Caching," IEEE/ACM Transactions on Networking, V. 10, N. 5, (October 2002).
This paper presented a detailed analysis of traces of DNS and its associated traffic at 2 locations -- MIT LCS and KAIST. This analysis resulted in insights into how various clients interact with the DNS, effects of caching and the performance perceived by the clients. It was widely believed that 2 factors contributed to scalability of DNS:
- Hierarchical Design around administratively delegated namespaces.
- Aggressive use of caching.
Caching was effective only when the TTLs were high as otherwise the cache will return stale values. However, low TTLs is not a very desirable thing to have in say for example mobile servers. This led the authors to delve deeper into estimating the effectiveness of DNS caching by observing the amound of DNS traffic in the wide area traffic. The following were some of the interesting things that were uncovered during these experiments:
- Low TTLs on A records (IP-hostname) mappings doesn't really affect performance due to the fact that network bandwidth is not really affected by it in practice. Further, caching at NS-record level results in substantial increase in DNS performance.
- The major cause of latency observed by the client is due to recursive referrals. For lookups involving 2 or more referrals, more than 95% took more than 100ms and 50% take more than 1 second to resolve.
- Almost half of the lookups were not associated with a TCP connection as was assumed initially.
No comments:
Post a Comment