Database Replication for Optimal Performance in Parallel BLAST

Parallelization of BLAST at the software level usually segments either the query or the database but not both. In this paper, we investigate a hybrid segmentation approach that combines database segmentation with query segmentation. By organizing the nodes into groups, splitting the queries among groups, and replicating the entire database at each group, we take advantage of both database segmentation and query segmentation. With an appropriate number of nodes in each group, the portion of the database at each node is small enough to reside in core memory, thus avoiding being paged to disks, and the intercommunication and synchronization between the workers and the leader of a group is contained, thus resulting a better performance.