Reducing DNS caching

Motivated by our ongoing work exploring an alternative Internet architecture, we wish to make use of naming services in order to support functionality such as: host and network mobility; application and/or virtual machine migration; and various forms of traffic control (e.g. multi-homing). Currently, the Domain Name System (DNS) is used to resolve names to DNS records, with relatively large time-to-live (TTL) values (several thousands of seconds) for caching the results. To support new agile services and systems, cached results may need to have much lower TTL values, so that cached DNS values do not become stale as system changes occur, e.g. changes to end-system location information to support new methods of mobility. However, current conventions for DNS configuration normally use conservatively high TTL values. We have conducted an empirical study of a live DNS deployment where we have reduced to zero the TTL values of records for the entire School of Computer Science at the University of St Andrews. Our results show that the increase in DNS load is much lower than might be expected, following a highly non-linear decrease with respect to the TTL value of the DNS records.