November 2016
Intermediate to advanced
576 pages
18h 22m
English
Juan Gómez-Luna and Izzat El Hajj
This chapter presents the parallel graph search pattern. Since graph search computation is about examining the vertex values, there is very little computation on these values once they are loaded from memory. As a result, the speed of graph search is typically limited by memory bandwidth. This chapter presents a graph data format similar to the CSR storage format for sparse matrix that helps minimize the consumption of memory bandwidth. We then introduce work queues, an important class of parallel data structures that supports work-efficient iterative algorithms that require dynamic discovery and collection of the data to be processed. We will show that the privatization ...