Home

GraphBrew Wiki

GraphBrew is a graph reordering framework. It composes lightweight primitives — Leiden communities, Rabbit Order, RCM, degree bucketing — into ten variants that match Gorder's cache quality at a fraction of the reorder cost.

Documentation

Start here

• Getting-Started — build, run your first benchmark
• Reordering-Algorithms — every algorithm explained
• Running-Benchmarks — command-line workflow

Reference

• Command-Line-Reference — all flags
• Supported-Graph-Formats — .sg, .el, .wel
• Graph-Benchmarks — graph catalog
• Troubleshooting — common errors and fixes
• FAQ — short answers to common questions

Deep dives

• GraphBrewOrder — the composable pipeline
• Cache-Simulation — bench/bin_sim/* usage
• ECG-Final-Runs — current ECG/gem5 final-run profiles, charged P-OPT, and supported baselines
• ECG-Sniper-Runs — Sniper backend setup, bounded SIFT full-wrapper validation, DROPLET, and current multithread blocker
• ECG-Slurm-Runs — split ECG/gem5 final runs across UVA Slurm and aggregate later
• Code-Architecture — codebase map
• VLDB-Experiments — reproducing the paper

Developer

• Contributing — adding algorithms and benchmarks
• Python-Scripts — analysis tools

Research-only (not part of the VLDB submission)

• AdaptiveOrder-ML — runtime algorithm selector

What GraphBrew gives you

Pipeline stage	Choices	What it controls
Community detection	Leiden, Rabbit Order	spatial locality
Intra-community ordering	BFS, RCM, HubCluster, DBG, Gorder	temporal locality
Inter-community arrangement	hierarchical sort, Rabbit on super-graph, RCM, tile	global layout

Variants ship as flags: -o 12:leiden, -o 12:rabbit, -o 12:hrab, -o 12:tqr, -o 12:hcache, -o 12:rcm, -o 12:hubcluster, -o 12:streaming.

See Reordering-Algorithms for the full list.

Repository

• Code: https://github.com/UVA-LavaLab/GraphBrew
• Issues: https://github.com/UVA-LavaLab/GraphBrew/issues
• Paper: see paper/main.tex (VLDB 2026 submission)