HPC Resources

Graph formats, converters, datasets and tools for high-performance parallel and distributed graph algorithms.

Graph Formats

Standard representations used across our tools and benchmarks. Full specifications at the Graph Formats documentation site.

Format	Type	Node IDs	Description
BGR	Binary (CSR)	0-indexed	Binary Graph Representation — compact CSR with adaptive uint32/uint64, parallel I/O, weighted/unweighted support
MTX	Text (Matrix Market)	1-indexed	Widely used text format for sparse matrices — human-readable edge list with header metadata
BVGraph	Compressed binary	0-indexed	WebGraph’s compressed format — gamma/delta/zeta coded successor lists, used by LAW datasets
ECL	Binary (CSR)	0-indexed	ECL graph format — simple binary CSR used by ECL-MIS, ECL-CC, ECL-BCC and related GPU algorithms
WGBin	Binary	0-indexed	WebGraph binary — intermediate raw format for BVGraph decoding pipelines

Graph Format Converters

The graph-format-converters repo converts WebGraph BVGraph compressed graphs (.graph + .properties) from the LAW dataset collection into standard formats.

Tools

Tool	Description
`bvgraph_gen_offsets`	Generate `.offsets` from `.graph` + `.properties` (pure C++, no Java)
`bvgraph_to_bgr`	Convert BVGraph → BGR (binary CSR)
`bvgraph_to_mtx`	Convert BVGraph → MTX (Matrix Market text)

Quick Start

# Build C++ tools (no Java required)
make cpp

# Download a graph (example: eu-2005, 862K nodes, 19M edges)
mkdir -p data
wget -P data http://data.law.di.unimi.it/webdata/eu-2005/eu-2005.graph
wget -P data http://data.law.di.unimi.it/webdata/eu-2005/eu-2005.properties

# Generate offsets, then convert to BGR
./bvgraph_gen_offsets data/eu-2005
./bvgraph_to_bgr data/eu-2005 data/eu-2005.bgr 8

Performance

Multi-threaded C++ (4 threads) converts BGR at ~150M edges/s.

Pipeline	MTX time	BGR time
C++	0.68s	0.26s
Java	1.76s	0.52s
Speedup	2.6×	2.0×

Benchmarked on eu-2005 (862K nodes, 19M edges), single thread.

Graph Datasets

Browse our curated collection of real-world and synthetic graphs:

Large Graphs — Graphs with 500M+ edges (web crawls, social networks, genomics, up to 2.5 trillion edges)
Sources — Where to find graph datasets (SNAP, LAW, SuiteSparse, KONECT, BioGraphs, and more)

Repository	Description
Graph-Formats	Documentation site for graph format specifications (BGR, MTX, BVGraph, ECL, WGBin)
graph-format-converters	C++ and Java tools to convert WebGraph BVGraph → BGR / MTX
SCC Analysis	Benchmark of 4 parallel SCC algorithms on graphs up to 92 billion edges
BCC-Finding	Biconnected components algorithms
PHEM	Parallel heterogeneous graph algorithms