edriver-doc.txt

it is very important

$Id: edriver-doc.txt,v 1.1 1996/10/04 13:28:51 ewz Exp $

Name: edriver

Function: Driver program to call evaluation routines. 

Usage:  edriver <filestem> [-nd] [-<f0><f1>]* 

Details:

    A set of metrics are computed by edriver and stored into
    a set of files.  The metrics are divided into two categories:
    scalar metrics and distribution metrics.  The scalar metrics
    are all stored in the file "./<filestem>-ev".  They are:

	average node degree
	diameter of graph (using each <f0><f1> combination (see below))
	average depth (using each <f0><f1> combination (see below))
	number of biconnected components

    The distribution metrics and filenames are as follows:

	node degree distribution ("./<filestem>-dd")
	depth distribution using <f0><f1> ("./<filestem>-<f0><f1>d")

    The purpose of the <f0><f1> specification(s) are to allow 
    computation of metrics using one or more edge weight fields.
    The edges in a graph may have up to three integer valued
    weights.  These weights are stored in the SGB arc fields
    named "len", "a" and "b".  We use the "len" field to store
    the Euclidean length of the arc; we use the "a" field in
    the hierarchical models to store a routing policy weight.
    The "b" field is currently not used in any of the graph models.
    Each edge can also be assumed to have unit weight.

    
    For each -<f0><f1> pair, shortest paths will be computed using 
    <f0> as the distance metric and <f1> as the measurement metric, 
    where <f0> and <f1> are one of {l, a, b, h}, indicating the arc 
    field to use (l = len, a = a, b = b) or to use unit weight (h).  
    The scalar properties of diameter and average depth will be stored 
    in the scalar file; if the -nd option is not present, the depth 
    distribution will be stored in the file <filestem>-<f0><f1>d.


    The file "./<filestem>.gb" is expected to contain a
    graph in SGB format, as created by gb_save().

    If the -nd option is given, only the scalar characteristics
    will be computed, and the distribution files will not
    created.  They take up lots of space for larger graphs.
    (Though it probably shouldn't be) the default is to evaluate 
    the distributions. 

    
Example usage:  edriver r10-0 -nd -hh -ll -hl 

     This will evaluate the graph in ./r10-0.gb without creating
     the distribution files.  Three types of shortest path metrics
     will be evaluated, corresponding to hops (hh), length (ll) and
     hop-computed-routes with length-computed-metrics (hl).
     Only one file will be produced, namely ./r10-0-0-ev.
     (The output file can be found in sample-graphs/rand/r10.)