Skip to content

cherryunix/Grail-

Repository files navigation

***************************************************************
*****************      GRAIL (improved)     *******************
**** Author :           Hilmi YILDIRIM      *******************
**** please contact yildih2@cs.rpi.edu for any problems *******
***************************************************************

The code is associated with the following papers:
1) Grail: scalable reachability index for large graphs ( VLDB'10 paper)
and
2) GRAIL: A Scalable Index for Reachability Queries in Very Large Graphs (journal version, under review)

Detailed usage is explained below, here we give sample usage
for the main methods of the papers above.  
---------------------------------------------------------------
1) ./grail sample.gra -test sample.test -dim 2
2) ./grail sample.gra -test sample.test -dim 2 -t -2 -ltype 1
---------------------------------------------------------------

This package is provided for the repeatability purposes. It may
contain some (experimental )sections that is not mentioned in 
the paper. PLEASE READ THE INSTRUCTIONS CAREFULLY
to run the program flawlessly. Feel free to use it for
research purposes.

  
The code is written in c++ and built on the codebase
of the paper PathTree (by Jin et al.). It uses the
same graph format.

To compile :
make

To run :
./grail -h

Usage:
 grail [-h] <filename> -test <testfilename> [-dim <DIM>] [-skip] [-ex] [-ltype <labelingtype>] [-t <alg_type>]
Description:
        -h                      Print the help message.
  <filename> is the name of the input graph in gra format.
        -test           Set the queryfile which contains a line of <source> <target> <reachability> for each query. 
        -dim            Set the number of traversals to be performed. The default value is 5.
        -ex             Use exception lists method instead of pruning. Default is not using exception lists.
        -skip           Skip the phase converting the graph into a dag. Use only when the input is already a DAG.
        -t <alg_type>           alg_type can take 8 different values.  Default value is 1.
                         1  : Basic Search (used in the original VLDB'10 paper) 
                         2  : Basic Search + Level Filter 
                         3  : Bidirectional Search 
                         6  : Bidirectional Search + Level Filter 
                         -1 : Positive Cut + Basic Search
                         -2 : Positive Cut + Basic Search + Level Filter (usually provides the best query times) 
                         -3 : Positive Cut + Bidirectional Search 
                         -6 : Positive Cut + Bidirectional Search + Level Filter
        -ltype <labeling_type>          labeling_type can take 6 different values.  Default value is 0.
                         0  : Completely randomized traversals.  
                         1  : Randomized Reverse Pairs Traversals (usually provides the best quality) 
                         2  : Heuristicly Guided Traversal: Maximum Volume First 
                         3  : Heuristicly Guided Traversal: Maximum of Minimum Interval First 
                         4  : Heuristicly Guided Traversal: Maximum Adjusted Volume First 
                         5  : Heuristicly Guided Traversal: Maximum of Adjusted Minimum Volume First 


INPUT GRAPH FORMAT: (see sample.gra)
 Notes : Second line gives the number of nodes n. The node ids should be in [0,n-1]. 

TEST FILE FORMAT: (see sample.test)
 - Each line contains triples (<source> <target> <reachability>)
 The program compares its output with <reachability> value of each query and prints a success ratio which is supposed to be 1.
 However if you dont know the ground truth, (i.e. for very large graphs) 
     you can give any value for <reachability> and ignore the success ratio.

 
GRAPH is already a DAG:
In this case please use -skip option to avoid recomputation of strongly connected components.

Selection of Dimensionality:
For sparse graphs. (avg. degree up to 3) -dim 2 is quite optimal.
For denser graphs, you can increase it up to 5. 

Selection of Labeling Type:
The best labeling type in our experiments is reverse pairs. (-ltype 1)
Please see the paper for details of different labelings.

Selection of the Algorithm:
The best algorithm in our experiments is (-t -2)
Please see the paper for details of different algorithms.

Sample usage:
./grail input.gra -test test.txt -t -2 -ltype 1 -dim 2 (this the main final algorithm proposed in 2nd paper)
./grail input.gra -test test.txt -t 1 -ltype 0 -dim 2 (this is the algorithm proposed in VLDB'10 paper )


The program reports the success rate, construction time, query time and the memory usage. 


About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published