As databases grow larger, the bulk loading feature that loads vast amount of data into the database becomes more critical. Bulk loading feature is classified into two categories: (1) data file bulk loading that loads the data into the data file by batch processing and (2) index bulk loading that generates the index by batch processing.
In this thesis, we design and implement bulk loading for the ODYSSEUS/ COSMOS object storage system, which has been developed at the Database and Multimedia Laboratory of KAIST. The characteristics of our bulk loading technique are as follows: (1) it takes advantage of physical contiguity by storing multiple pages with one I/O call; (2) it expedites data clustering in data file bulk loading by piplining sorting and loading;(3) it supports incremental bulk loading of an existing $B^+$ tree index; (4) it can precisely control the space utilization.
