Proactive caching of DNS records: addressing a performance bottleneck

Abstract

The resolution of a host name to an IP-address is a necessary predecessor to connection establishment and HTTP exchanges. Nonetheless, DNS resolutions often involve multiple remote name-servers and prolong Web response times. To alleviate this problem name servers and Web browsers cache query results. Name servers currently incorporate passive cache management where records are brought into the cache only as a result of clients' requests and are used for the TTL duration (a TTL value is provided with each record). We propose and evaluate different enhancements to passive caching that reduce the fraction of HTTP connection establishments that are delayed by long DNS resolutions. Renewal policies refresh selected expired cached entries by issuing unsolicited queries. Trace-based simulations using Web proxy logs demonstrated that a significant fraction of cache misses can be eliminated with a moderate overhead. Simultaneous-validation (SV) transparently uses expired records. A DNS query is issued if the respective cached entry is no longer fresh, but concurrently, the expired entry is used to connect to the Web server and fetch the requested content. The content is served only if the expired records used turn out to be in agreement with the query response.