External memory graph algorithms and applications to geographic information systems

• Authors: Laura Iulia Toma;Lars Arge
• Venue: External memory graph algorithms and applications to geographic information systems
• Year: 2003

Many modern applications need to process data that is too large to fit in the internal memory of the computer and must be stored on disk. Since disks are much slower than CPU, the bottleneck in many such applications is the I/O-communication between internal memory and disk. One approach to improve performance is to design algorithms that optimize the I/O communication between internal memory and disk, known as external memory algorithms or I /O-efficient algorithms. This thesis focuses on external memory algorithms for basic graph problems and the application of the principles of external memory computation to Geographic Information Systems (GIS). The first part of the thesis presents external memory algorithms for a number of graph problems including breadth-first search (BFS), depth-first search (DFS) and computation of shortest paths. Most of the graph theoretical results in this thesis are for planar graphs. We have initiated a sequence of I/O-efficient reductions between planar BFS, DFS, single-source shortest path (SSSP) and graph separation which led to optimal algorithms for these problems on planar graphs. We have extended the techniques to planar directed graphs obtaining algorithms for BFS, SSSP, all-pair-shortest-paths (APSP) and topological sort and developed a data structure for shortest path queries. The second part of the thesis presents I/O-efficient algorithms for flow modeling on grid-based terrains and the system that we designed and implemented, TERRAFLOW. We present experimental result that demonstrate the scalability our system. On large terrains, TERRAFLOW greatly outperforms existing systems. TERRAFLOW has been recognized in the GIS community and is being incorporated into both major GIS systems, GRASS and ArcInfo.