When we did project 3 it was hard to see the advantages of bottom-up bfs because probably the test graphs were probably unable to demonstrate well the situations in which top-down bfs does not perform as well. However, the paper brings up some important points about how bottom-up reduces the number of atomic operations required and also traverses fewer edges in the case of large frontiers.
"Direction-Optimizing Breadth-First Search." Beamer et al. Supercomputing 2012.
http://parlab.eecs.berkeley.edu/sites/all/parlab/files/main.pdf
When we did project 3 it was hard to see the advantages of bottom-up bfs because probably the test graphs were probably unable to demonstrate well the situations in which top-down bfs does not perform as well. However, the paper brings up some important points about how bottom-up reduces the number of atomic operations required and also traverses fewer edges in the case of large frontiers.