A study of the reliability of Internet sites

Failure and repair rates of components are often assumed to be exponentially distributed. This hypothesis is testable for failure rates, though the process of gathering and reducing the data to a usable form can be difficult. By applying an appropriate test statistic, some samples were found to have a realistic change of being drawn from an exponential distribution, while others can be confidently classed as nonexponential. Data were collected from a large number of hosts via the Internet. Almost all of the visible Internet (over 350000 hosts) were considered, and more than 68000 of these that were judged likely to respond were queried. These hosts were sampled several times to obtain up-times, and finally to determine average host availability. Estimates of availability, mean-time-to-failure, and mean-time-to-repair were derived. The results reported correspond with those commonly seen in practice