Abstract: A parallel query manager accepts a list of file extents to be searched and produces a number of search lists, one for each disk to be searched. The query manager first uses a mapper to find out how the database spaces are stored on disk. It then matches the search extent list with the mapping information to determine which parts of which disks are to be searched. It then initiates several searches in parallel so that all the affected disks can be kept busy at the same time. The query manager then checks for return data on each stream, and merges the results.

Abstract: A database system holds data in the form of a sequence of records, each record comprising one or more fields. The database can be interrogated by a search query, which specifies a particular logical combination of comparisons to be performed on specified fields of each record. Before the search commences, the search query is compiled to produce an optimized sequence of code for search within a record. Each comparison operation is assigned a cost, reflecting the cost in time to retrieve the required fields and to perform the comparisons, and is also assigned a probability, indicating the probability that the comparison will produce a true result. Each logical operation in the search query is then processed, to find the order of handling its arguments that gives the minimum expected cost, and the arguments are re-arranged into that order.