Scientific and reference databases on the Web are a primary source of information. However, the data is subject to continuous change and often only the most recent versions are preserved. For many database providers it has become common practice to overwrite existing database states when changes occur and regularly publish new releases of the data on the Web. Failure to archive earlier states of the data may lead to the loss of scientific evidence, and the basis of findings may no longer be verifiable. Recently, some database providers have initiated archiving efforts that allow to view entries exactly as they were in the past.

Database archiving is, however, not only important for verification of scientific findings. Consider the fact that nearly every dictionary, gazetteer, encyclopaedia or reference manual that one traditionally found on the reference shelves of libraries is now available on the Web. In many cases one is interested not just in an old version of the data, but one is interested in forming queries over the history of the data.

A prime example of this is the CIA World Factbook, possibly the most widely used source of demographic information. Over the past 18 years the Factbook has moved from an annually printed distribution to a web resource. Over that period it has kept a remarkably uniform structure. Apart from one or two spreadsheets that contain only a small portion of the data, there appears to have been no attempt to bring all past versions into a common form. In fact, although all past versions are available on line (the printed versions have been transcribed into a variety human and machine-readable formats), it is quite hard even to find all these versions. Temporal queries such as ''How did electricity generation in China change over the past 15 years?'' can be extremely valuable, but to answer such queries using the data in its current state is extremely time-consuming. This emphasises the need to have both a uniform archival representation and a query language in order to make temporal queries like this easy to formulate and efficient to execute.

A system trying to archive evolving databases on the Web faces several challenges. First and foremost, the systems needs to be able to efficiently maintain and query multiple snapshots of ever growing databases. Second, the system needs to be flexible enough to account for changes to the database structure and to handle data of varying quality. Third, the system needs to be robust and invulnerable to local failure to allow reliable long-term preservation of archived information. Our archive management system XArch addresses the first challenge by providing the functionality to maintain, populate, and query archives of database snapshots in hierarchical format. In our ongoing efforts we are currently improving XArch regarding archiving evolving databases, and databases of varying quality.

The following poster gives an overview of the techniques and architecture behind XArch.