Bulk-Parallel Priority Queue in External Memory
Thomas Keh · Repository KITopen (Karlsruhe Institute of Technology) · 2014
We present a priority queue implementation with support for external memory. The focus of our work has been to derive a benefit from parallel shared-memory machines. It's the first parallel optimization of an external-memory priority queue. An additional bulk insertion interface accelerates longer sequences of homogeneous operations, as they are more likely to occur in applications that process large amounts of data. The algorithm will be available as an extension to the STXXL, a popular C++ template library for extra large data sets. Experiments have shown great improvements over the current external-memory priority queue of the STXXL for homogeneous bulk operations. However, the high overhead for spawning threads, as well as the need for cache synchronization in the global ExtractMin operation, show the inherent limitations of the parallelizability of priority queues.