Homework Assignment #4

Due: Tuesday, November 27

---------------------------------------------------------------------
Reading assignments (available from the class website):

1) A Theoretician's Guide to the Experimental Analysis of Algorithms,
   D.S. Johnson, 2002.  
2) Reach for A*, Efficient Point-to-Point Shortest Path Algorithms,
   A.V. Goldberg, H. Kaplan, and R.F. Werneck, 2006.
---------------------------------------------------------------------

For this homework assignment, you may work individually or form small
groups of students.

Shortest path problems are ones of the most fundamental combinatorial
optimization problems with many applications, both direct and as
subroutines in other combinatorial optimization algorithms. Algorithms
for these problems have been studied since 1950's and still remain an
active area of research.

Our class is going to hold a mini-competition based on the 
9th DIMACS Implementation Challenge on Shortest Paths:

   http://www.dis.uniroma1.it/~challenge9/

Your challenge is to answer queries of point-to-point shortest paths
in the USA Road graph, the (undirected) road networks of the 50 US
States and the District of Columbia, with around 24 million nodes and
29 million edges.  We will use ROAD DISTANCE as the weight function on
each edge.

Two input data files are available locally on anonymous ftp:

   ftp://ftp.cc.gatech.edu/pub/people/bader/CSE6140/USA-road-d.USA.gr.gz 
   ftp://ftp.cc.gatech.edu/pub/people/bader/CSE6140/USA-road-d.USA.co.gz 

They are in ".gr" and ".co" formats, described at 

   http://www.dis.uniroma1.it/~challenge9/format.shtml

If you want more information on the data source, see

   http://www.dis.uniroma1.it/~challenge9/data/tiger/

You may design and implement any algorithm, sequential or parallel,
that solves this problem of answering point-to-point distance queries
between a pair of nodes. You may also use any algorithms and source
code legitimately available from other researchers, optimize these
codes, or design new algorithms.

An input file of 1,000 random queries is available (in ".p2p" format)
from anonymous ftp:

   ftp://ftp.cc.gatech.edu/pub/people/bader/CSE6140/USA-road-d.USA.p2p

Note that you will receive a new set of 1,000 random queries in the
same format to use for the mini-competition. The file will be made available
during class on November 27 at:

   ftp://ftp.cc.gatech.edu/pub/people/bader/CSE6140/USA-road-d.USA.v2.p2p

You will report the average time for your implementation to find the
DISTANCE between each pair of vertices in the query. (You do not need
to report the actual path.)

You may also pre-process the road graph before handling the
queries. Please report this pre-processing time.  Report technical
data for the machine on which queries were performed in the format
described.

You may have correctness checking in you code. Correctness checking
time should not count in the computation time.

A reference implementation with DIMACS format parser is available from:

   ftp://ftp.cc.gatech.edu/pub/people/bader/CSE6140/mlb-dimacs.tar.gz 

--------------------------------------------

You must submit a gzipped-tarfile of your source code to me by 3PM on
November 27.

I will have the winning teams present an overview of their
algorithm/approach, and give performance results.  The team with the
best performance will win a prize!

Good luck!