dreambydream / rabbit_order

Rabbit Order

• This is an implementation of the algorithm proposed in the following paper:
  • J. Arai, H. Shiokawa, T. Yamamuro, M. Onizuka, and S. Iwamura. Rabbit Order: Just-in-time Parallel Reordering for Fast Graph Analysis. IEEE International Parallel and Distributed Processing Symposium (IPDPS), 2016.
• Please read license.txt before reading or using the files.
• Note that some graph datasets are already reordered, and so Rabbit Order will not show significant performance improvement on those graphs.
  • For example, Laboratory for Web Algorithmics reorders graphs using the Layered Label Propagation algorithm.
  • Web graphs are sometimes reordered by URL. This ordering is known to show good locality.

How to use

demo/reorder.cc is a sample program that reorders graphs by using Rabbit Order. Type make in the demo/ directory to build the program.

Requirements

• g++ (4.9.2)
• Boost C++ library (1.58.0): boost install
• libnuma (2.0.9): sudo apt install libnuma-dev
• libtcmalloc_minimal in google-perftools (2.1): https://blog.csdn.net/woaiwojia6699/article/details/111224528

Numbers in each parenthesis are the oldest versions that we used to test this program.

Usage

Usage: reorder [-c] GRAPH_FILE
  -c    Print community IDs instead of a new ordering

If flag -c is given, this program runs in the clustering mode (described later); otherwise, it runs in the reordering mode.

Input

GRAPH_FILE has to be an edge-list file like the following:

14 10
2 194
14 1
89 20
...

Each line represents an edge. The first number is a source vertex ID, and the second number is a target vertex ID. Edges do not need to be sorted in some orderings, but vertex IDs should be zero-based contiguous numbers (i.e., 0, 1, 2, ...); otherwise, this demo program may consume more memory and show lower performance.

Many edge-list files in this format are found in [Stanford Large Network Dataset Collection] (http://snap.stanford.edu/data/index.html).

Output (reordering mode)

The program prints a new ordering as follows:

8
16
1
4
...

These lines represent the following permutation:

Vertex 0 ==> Vertex 8
Vertex 1 ==> Vertex 16
Vertex 2 ==> Vertex 1
Vertex 3 ==> Vertex 4
...

Output (clustering mode)

The program prints a clustering result as follows:

1
5
1
5
5
9
...

These lines represent the following classification:

Vertex 0 ==> Cluster 1
Vertex 1 ==> Cluster 5
Vertex 2 ==> Cluster 1
Vertex 3 ==> Cluster 5
Vertex 4 ==> Cluster 5
Vertex 5 ==> Cluster 9
...

Note that the cluster IDs may be non-contiguous.