Supported Graph Formats

Supported Graph Formats

Five formats are supported. .sg is the binary form GraphBrew converts everything to on first load; subsequent runs use the cached .sg.

Quick Reference

Format	Extension	Description	Weighted
Edge List	.el	Simple text format	No
Weighted Edge List	.wel	Text with weights	Yes
Matrix Market	.mtx	Standard sparse matrix	Optional
DIMACS	.gr	Challenge format	Yes
Serialized	.sg, .graph	Binary format	Optional

---

Edge List Format (.el)

Description

The simplest format: one edge per line with source and destination vertices.

Format

<source> <destination>
<source> <destination>
...

Example

0 1
0 2
1 2
2 3
3 4

Properties

• Vertices are 0-indexed
• Whitespace separated (space or tab)
• Directed by default (add -s to symmetrize)
• No header required

Usage

./bench/bin/pr -f graph.el -s -n 3

Converting From Other Formats

# From CSV
cat graph.csv | tr ',' ' ' > graph.el

# Remove headers
tail -n +2 graph_with_header.el > graph.el

---

Weighted Edge List Format (.wel)

Description

Edge list with weights for SSSP and weighted algorithms.

Format

<source> <destination> <weight>
<source> <destination> <weight>
...

Example

0 1 1.5
0 2 2.0
1 2 0.5
2 3 1.0

Properties

• Weights are floating-point numbers
• Used for SSSP benchmark
• Weights can be integers or decimals

Usage

./bench/bin/sssp -f graph.wel -s -r 0 -n 3

Creating Weighted Graphs

# Add random weights to edge list
awk '{print $1, $2, rand()}' graph.el > graph.wel

# Add unit weights
awk '{print $1, $2, 1.0}' graph.el > graph.wel

---

Matrix Market Format (.mtx)

Description

Standard format from the Matrix Market collection. Widely used in scientific computing.

Format

%%MatrixMarket matrix coordinate pattern general
<rows> <cols> <nnz>
<row> <col>
<row> <col>
...

Or with values:

%%MatrixMarket matrix coordinate real general
<rows> <cols> <nnz>
<row> <col> <value>
...

Example

%%MatrixMarket matrix coordinate pattern general
5 5 6
1 2
1 3
2 3
3 4
4 5
5 1

Properties

• 1-indexed (vertices start at 1, not 0)
• Header line describes matrix type
• Dimensions line: rows, columns, non-zeros
• pattern = unweighted, real = weighted
• symmetric = edges only stored once

Usage

./bench/bin/pr -f graph.mtx -s -n 3

Common Matrix Market Types

Type	Meaning
pattern	Binary (0/1), unweighted
real	Floating-point weights
integer	Integer weights
general	Full matrix (not symmetric)
symmetric	Only lower triangle stored

---

DIMACS Format (.gr)

Description

Format from DIMACS implementation challenges. Common for road networks.

Format

c Comment line
p sp <nodes> <edges>
a <source> <destination> <weight>
a <source> <destination> <weight>
...

Example

c Example graph
c This is a comment
p sp 5 6
a 1 2 10
a 1 3 20
a 2 3 5
a 3 4 15
a 4 5 10
a 5 1 25

Properties

• 1-indexed vertices
• c lines are comments
• p sp defines problem (shortest path)
• a lines are arcs (edges)
• Always weighted

Usage

./bench/bin/sssp -f graph.gr -s -r 0 -n 3

DIMACS Sources

• DIMACS Challenge
• Road networks (USA, Europe)

---

Serialized Binary Format (.sg, .graph)

Description

Binary format for fast loading. Pre-computed CSR representation.

Creating Serialized Graphs

# Convert edge list to serialized format
./bench/bin/converter -f graph.el -s -b graph.sg

Usage

./bench/bin/pr -f graph.sg -n 3

Properties

• Much faster loading for large graphs
• Larger file size than text
• Architecture-specific (endianness)
• Includes symmetrization if applied

When to Use

Graph Size	Recommendation
< 1M edges	Use text format
1M - 100M edges	Consider serialized
> 100M edges	Use serialized

---

Format Detection

GraphBrew automatically detects format by extension:

Extension	Detected Format
.el	Edge list
.wel	Weighted edge list
.mtx	Matrix Market
.gr	DIMACS
.sg, .graph	Serialized

Important Notes

• Format detection is based solely on file extension
• There are no flags to force a specific input format
• Rename files or convert them if extension doesn't match format

---

Converting Between Formats

Using the Converter

# Edge list to serialized
./bench/bin/converter -f graph.el -s -b graph.sg

# Edge list to MTX format
./bench/bin/converter -f graph.el -s -p graph.mtx

# Edge list to Ligra format
./bench/bin/converter -f graph.el -s -y graph.ligra

# MTX to edge list (manual)
tail -n +3 graph.mtx | awk '{print $1-1, $2-1}' > graph.el

Python Conversion

import networkx as nx

# Read various formats
G = nx.read_edgelist('graph.el', nodetype=int)
G = nx.read_weighted_edgelist('graph.wel', nodetype=int)

# Write to edge list
nx.write_edgelist(G, 'output.el', data=False)

Common Conversions

# MTX to EL (1-indexed to 0-indexed)
grep -v "^%" graph.mtx | tail -n +2 | awk '{print $1-1, $2-1}' > graph.el

# GR to WEL
grep "^a" graph.gr | awk '{print $2-1, $3-1, $4}' > graph.wel

# CSV to EL
cat graph.csv | tr ',' ' ' | grep -v "source" > graph.el

---

Downloading Graphs

Use the unified pipeline to download graphs automatically:

python3 scripts/graphbrew_experiment.py --download-only --size small

Manual sources: SNAP, SuiteSparse, Network Repository. See Benchmark-Suite for size categories.

---

Graph Properties

Required Properties

Property	Requirement
Vertices	Non-negative integers
Self-loops	Allowed (removed by default, use -S to keep)
Multi-edges	Allowed (may affect results)
Directed	Yes (use -s for undirected)

Recommended Properties

Property	Recommendation
Connected	Ideally connected or mostly connected
No isolates	Remove isolated vertices for cleaner results
Reasonable size	100+ edges for meaningful benchmarks

Validating Graphs

# Check basic properties
head -5 graph.el
wc -l graph.el  # Edge count

# Check for issues
sort graph.el | uniq -c | sort -rn | head  # Duplicate edges
awk '$1==$2' graph.el  # Self-loops

---

Troubleshooting

Common issues: Windows line endings (dos2unix), 1-indexed vertices (awk '{print $1-1, $2-1}'), empty files. See Troubleshooting for details.

---

Test Graphs

GraphBrew includes a test graph in scripts/test/graphs/:

File	Description
tiny/tiny.el	Tiny edge list for quick testing

Use for quick testing:

./bench/bin/pr -f scripts/test/graphs/tiny/tiny.el -s -n 3

---

Next Steps

• Running-Benchmarks - How to run benchmarks
• Getting-Started - Getting started guide
• Graph-Benchmarks - Available algorithms

---

← Back to Home | Running Benchmarks →