The NETWORK Procedure

This example looks at the road networks in the state of New York (NY). The distance graph raw data are maintained at the DIMACS challenge website (Demetrescu 2010). The NY road network includes 264,346 intersections (nodes) and 733,846 roads (links). Although the input data table is not large, the computing power that you need to find all-pairs shortest paths is enormous. In addition, the storage space that you need to handle the results data can easily overwhelm the capacity of a single machine. In this example, a session of 100 machines (each with 32 cores) was configured to process this graph.

The following statements construct the links data table mycas.RoadNY from a local copy of the raw distance graph data:

filename in 'USA-road-d.NY.gr';
data mycas.RoadNY (drop=a);
   infile in firstobs=8;
   input a $ from $ to $ weight;
run;

The following statements find the all-pairs shortest paths of the NY road network (that have a total path weight of less than 20,000) by using a distributed algorithm. This algorithm takes advantage of all the machines and cores in your configured session.

proc network
   logFreqTime      = 10
   logLevel         = aggressive
   direction        = directed
   links            = mycas.RoadNY;
   shortestPath
      maxPathWeight = 20000
      outWeights    = mycas.shortPathSummary
      outPaths      = mycas.shortPathPaths;
run;
%put &_NETWORK_;

The progress of the procedure is shown in Output 2.13.1.

Notice that the resulting output data tables, mycas.shortPathSummary and mycas.shortPathPaths, are large distributed data tables.