[pgrouting-dev] [SoC][pgRouting] GSoC 2018 Final Report : Implement bellman-ford and parallel dijkstra for pgRouting.

pgRouting pgrouting-dev
1

Hello all,

I am Sourabh Garg, and This is my final report for my GSoC project.

Title: Implement Parallel Dijkstra’s and Bellman-Ford algorithm by the Boost Graph Library.

Organization: PgRouting under OSGeo

Abstract:

Graph Algorithms like Dijkstra’s single source shortest path algorithm is widely applied in many routing applications, but it is constrained by non-negative edge’s weight and is more efficient to use DAG shortest path algorithm(topological sort) in case of Directed Acyclic Graph.

Finding the shortest path from a source vertex to target using the bellman-ford algorithm for Graphs with negative weighted edges and dag_shortest_path algorithm for the Directed acyclic graph are major

deliverables.

Often such algorithms are applied over Large-scale graphs, where computation problem may arise. It may be beneficial to exploit the high-performance parallel computing system, by implementing distributed graph algorithms in pgRouting. I have developed the testing base for parallel Boost Graph algorithm.

Implemented functions

:
  • pgr_bellmanFord() - Returns the shortest path for the graph with negative weighted edges.
  • pgr_dagShortestPath() - Returns the shortest path for the Directed Acyclic Graph.

State of the art before the project: pgRouting didn’t have above functionalities before my GSoC.



Addition that my project brought to pgRouting: 



```
The deliverables are code, full documentation, documentation tests, pgTap of above functions.
```





**Pending Work:** I have done experimentation with boost parallel graph library to generate distributed graph and apply dijkstra's algorithm over it. It was quite challenging to integrate the parallel version of algorithms with pgRouting architecture.


Future Directions: 

  • Implement full functionality of parallel Dijkstra in pgrouting architecture.

  • Implement the feature to include the negative weighted edges in pgRouting graphs.

Links:

Best Regards,

Sourabh Garg