Mariposa (database)

Mariposa was a relational database research project run by Michael Stonebraker at UC Berkeley. Mariposa focused on creating wide-area distributed databases using an economic model in which querying servers "buy" data from data servers which "sell" it. The result is a network of servers that quickly reaches a best-performance point.

Other distributed RDBMSs rely on a heavyweight solution for moving data in which every server maintains a complete copy of the "master" database. While this works with private databases of small size, it simply cannot work with thousands of servers with aggregate data sets on the order of those being commonly found on the Internet.

Mariposa uses an entirely different paradigm for distribution in which each server includes only the data it needs, but maintains links to other servers and knows what data they contain. If the data needs of a particular server mean it is constantly "buying" data from another server, notably one with a high cost (say on the far side of the Atlantic), the servers will automatically copy the data over.

Mariposa's system is also largely automatic. Traditional systems require the dedicated resources of a database administrator in order to maintain performance as the data load changes over time, whereas in Mariposa the basic system will automatically handle such cases.

Unlike Stonebraker's earlier efforts which were groundbreaking, Mariposa seems to have had little effect on the marketplace as a whole to date. The project has since ended.

External links

• Mariposa