LDBC Graphalytics ArcadeDB Platform Driver

Platform driver implementation for the LDBC Graphalytics benchmark using ArcadeDB.

Uses ArcadeDB in embedded mode with the Graph Analytical View (GAV) engine, which builds a CSR (Compressed Sparse Row) adjacency index for high-performance graph algorithm execution with zero GC pressure.

This repository contains three benchmark modes:

1. Official LDBC Graphalytics — standardized framework with per-algorithm isolation, validation, and reporting
2. Native multi-vendor comparison — load once, run all algorithms, compare ArcadeDB vs Kuzu vs DuckPGQ vs Memgraph vs Neo4j vs FalkorDB vs HugeGraph
3. LSQB (Labelled Subgraph Query Benchmark) — 9 subgraph pattern matching queries on the LDBC SNB social network, comparing ArcadeDB (Cypher) vs DuckDB (SQL) vs FalkorDB (Cypher) and others

Supported Algorithms

Algorithm	Implementation	Complexity
BFS (Breadth-First Search)	Parallel frontier expansion with bitmap visited set and push/pull direction optimization	O(V + E)
PR (PageRank)	Pull-based parallel iteration via backward CSR	O(iterations * E)
WCC (Weakly Connected Components)	Synchronous parallel min-label propagation	O(diameter * E)
CDLP (Community Detection Label Propagation)	Synchronous parallel label propagation with sort-based mode finding	O(iterations * E * log(d))
LCC (Local Clustering Coefficient)	Parallel sorted-merge triangle counting	O(E * sqrt(E))
SSSP (Single Source Shortest Paths)	Dijkstra with binary min-heap on CSR + columnar weights	O((V + E) * log(V))

Prerequisites

• Java 21 or later (required for jdk.incubator.vector SIMD support). The published benchmark numbers use Eclipse Temurin 25 with -XX:+UseCompactObjectHeaders (GraalVM is not used)
• Maven 3.x
• Python 3.10+ (for Mode 2 and Mode 3 multi-vendor comparisons; see pyproject.toml for per-vendor install extras)

Build

mvn package -DskipTests

The build produces a self-contained distribution in graphalytics-1.3.0-arcadedb-0.1-SNAPSHOT/.

Dataset

Use the built-in dataset manager to browse and download datasets from the LDBC data repository:

# See all available datasets (40+ Graphalytics + 9 LSQB scale factors)
python3 datasets.py available

# Download the standard Graphalytics benchmark dataset (633K vertices, 34M edges, ~155 MB)
python3 datasets.py download datagen-7_5-fb

# Download the LSQB social network dataset (SF1, ~3.9M vertices, ~17.9M edges)
python3 datasets.py download lsqb-sf1

# Show downloaded datasets with size and vertex/edge counts
python3 datasets.py

Datasets are downloaded into the datasets/ directory (git-ignored). After downloading datagen-7_5-fb:

datasets/
  datagen-7_5-fb/
    datagen-7_5-fb.v              # vertex file (one ID per line)
    datagen-7_5-fb.e              # edge file (src dst weight, space-separated)
    datagen-7_5-fb.properties     # graph metadata
    datagen-7_5-fb-BFS/           # validation data per algorithm
    datagen-7_5-fb-WCC/
    ...

---

Weekly unattended run

python3 weekend.py --jvm-flags "-XX:+UseCompactObjectHeaders"

Runs every suite and vendor in sequence (ArcadeDB embedded Java benchmarks, the Graphalytics and LSQB multi-vendor suites, optionally Mode 1 with --mode1-dist) and writes a Markdown/JSON report to weekly-results/<timestamp>/. Each vendor runs in its own process group with a total-time and a no-output limit; a hung vendor gets SIGTERM, then SIGKILL, and the run continues. Loaded databases are kept under ~/.cache/ldbc-graph-bench and reused, so only the first run (or --reset, a new image, a changed dataset) pays for the loads. See CLAUDE.md for the details and python3 -m unittest discover -s tests for the tests.

Benchmarks and results

Every number is a warm number (first call discarded, median of 3 timed runs), measured one system at a time on AC power and validated against the official LDBC reference outputs (or the official expected counts for LSQB). A value marked ✗ failed that check and is not ranked; N/A = no implementation; timeout = the 5-minute limit per operation. Footnote symbols (§ ‡ ¶ *) are explained on the detailed pages. Machine: MacBook Pro 16" (2026), Apple M5 Pro, 48 GB RAM; ArcadeDB on Eclipse Temurin 25 with -XX:+UseCompactObjectHeaders, 12 GB heap for every JVM system, Docker Desktop with 32 GB.

Benchmark	What it measures	Detailed page
Graphalytics, official framework (Mode 1)	ArcadeDB only, official LDBC harness, one graph load per algorithm	docs/benchmark-graphalytics-official.md
Graphalytics, multi-vendor (Mode 2), datagen-7_5-fb	6 algorithms, 10 systems, 633K vertices / 34M edges	docs/benchmark-graphalytics-multivendor.md
Graphalytics, multi-vendor, graph500-22	5 algorithms, 10 systems, 2.4M vertices / 64M edges	docs/benchmark-graphalytics-graph500-22.md
LSQB SF1 (Mode 3)	9 subgraph pattern matching queries, 7+ systems	docs/benchmark-lsqb.md
ArcadeDB native image against the JVM Docker image	arcadedata/arcadedb:latest-native (GraalVM, no JVM) and a JVM image of the same engine commit on both Graphalytics graphs: compute equal to 1.6x slower, wall-clock slower, start-up 4.5x faster, 20-40% less peak memory	docs/benchmark-arcadedb-native-image.md

How ArcadeDB itself changes from release to release (official framework, Mode 2 and LSQB): ArcadeDB-release-progress.md.

Graphalytics, datagen-7_5-fb (633,432 vertices, 34,185,747 edges)

Seconds, engine-reported compute time of the algorithm (like the official Graphalytics processing time; ʷ = the engine reports none, wall-clock shown; the wall-clock client view and the rule are on the detailed page, re-measured 2026-10-07), last column peak memory in GiB. Bold marks the fastest valid result per column. Systems, versions, how we measure, per-system notes and how to run each vendor: detailed page.

System	Load	PageRank	WCC	BFS	LCC	SSSP	CDLP	Peak memory (GiB)
ArcadeDB embedded	71.9	0.085	0.003	0.020	2.05	0.75	0.96	5.6
ArcadeDB Docker	43.9	0.119	0.004	0.008	2.41	1.19	1.38	12.7
Neo4j	657	6.84§	0.021	0.45ʷ‡	15.4	N/A	N/A	13.3
Kuzu	28.8	0.85	0.181	0.043	N/A	N/A	N/A	0.8
LadybugDB	5.16	N/A	N/A	7.45	N/A	N/A	N/A	0.9
DuckPGQ	0.85	1.61✗	2.14♦	timeout¶	12.4	N/A	N/A	7.3
Memgraph	437	4.23	128	3.28	N/A	58.1	timeout	25.2
ArangoDB *	726	70.3	23.9	35.6	N/A	142	211✗	21.1
FalkorDB	116	0.96✗	0.96	0.059	N/A	N/A	6.96✗	8.3
HugeGraph	34.7	2.45	0.306	0.201	115	N/A	22.3✗	4.0

♦ DuckPGQ's WCC varies between 1.8 s and 8.0 s across runs on this machine (median of three runs shown).

Graphalytics, graph500-22 (2,396,657 vertices, 64,155,735 edges)

Seconds, engine-reported compute time (same rule as above; ʷ = wall-clock, ◊ = earlier-pass value); no SSSP (not defined for this dataset). ArcadeDB is 26.11.1-SNAPSHOT; bold marks the fastest valid result per column. Per-system notes, the harness problems found and how to reproduce: detailed page.

System	Load	PageRank	WCC	BFS	LCC	CDLP	Peak memory (GiB)
ArcadeDB embedded	114.5	0.268	0.013	0.076	46.9	1.78	1.6**
ArcadeDB Docker	101†	0.362	0.014	0.039	48.6	3.58	12.9
Neo4j	2017†	13.1	0.060	1.08ʷ	229ʷ	N/A	13.3
Kuzu	53.8†	3.30	0.854	0.161	N/A	N/A	1.6
LadybugDB	10.1†	N/A	N/A	16.6	N/A	N/A	1.7
DuckPGQ	0.57†	16.5✗	5.15	timeout‡	119	N/A	12.2
Memgraph	932§	19.0	OOM	10.2	N/A	timeout	30.3
ArangoDB 3.11.14	1700†	252ʷ◊	83.0	OOM¶	N/A	timeout	30.4
FalkorDB	348.5	2.19✗	2.03	0.163	N/A	20.8✗	15.7
HugeGraph	34.2†	7.08	0.706	0.418	timeout	22.6✗	7.8

OOM = out of memory (Docker Desktop's 32 GB). † remembered original load, ‡ DuckPGQ BFS does not finish (ended after 20 min), § Memgraph load includes the reverse-edge step, ¶ ArangoDB container ran out of memory, ** live heap after GC (process peak not measured): see the detailed page.

LSQB SF1 (3,947,829 vertices, 17,882,623 edges)

Seconds; ArcadeDB embedded is shown with the Graph Analytical View (OLAP) and without it (OLTP); bold marks the fastest result per query. All counts match the official expected output. Queries, run commands and analysis: detailed page.

System	Load	Q1	Q2	Q3	Q4	Q5	Q6	Q7	Q8	Q9	Peak GiB
ArcadeDB OLAP	119.5	0.08	0.13	0.05	0.04	0.17	0.07	0.04	0.11	0.30	4.3
ArcadeDB OLTP	159.2	2.45	4.40	3.17	1.06	11.39	11.00	1.07	6.91	0.87	12.0
ArcadeDB Docker	99.4	0.17	0.17	0.06	0.05	0.25	0.13	0.06	0.14	0.33	12.9
DuckDB	0.46	0.11	0.01	0.04	0.06	0.04	1.84	0.07	0.07	6.03	0.9
Kuzu	2.44	4.61	0.15	2.30	N/A	N/A	1.38	N/A	N/A	6.39	5.1
LadybugDB	2.95	0.12	0.10	10.44	0.17	0.18	0.66	0.41	0.32	0.07	8.8
Neo4j	252.0	4.99	1.63	10.76	6.28	5.66	28.49	8.04	12.39	254.29	13.3
PostgreSQL	15.1	9.17	0.65	1.60	6.35	2.17	15.50	11.16	3.39	50.62	0.5
Memgraph	222.2	58.06	timeout	timeout	4.30	3.54	121.15	4.92	3.02	timeout	3.1
FalkorDB	659.3	36.43	53.78	4.59	3.53	3.93	40.93	47.96	4.90	225.22	2.4

More

• docs/benchmark-graphalytics-official.md, docs/benchmark-graphalytics-multivendor.md, docs/benchmark-graphalytics-graph500-22.md, docs/benchmark-lsqb.md, docs/benchmark-arcadedb-native-image.md: one page per benchmark (how to run, method, full tables, notes).
• docs/vendor-notes.md: SurrealDB and Dgraph (excluded from default runs, see also SURREALDB.md) and FalkorDB.
• docs/architecture.md: file structure, the Graph Analytical View (CSR) engine, execution modes.
• ArcadeDB-release-progress.md: ArcadeDB across releases. results-*.md: raw results and history of earlier runs (latest: results-native-vs-jvm-2026-10-10.md). CLAUDE.md: benchmark rules for the harness.

License

Apache License, Version 2.0