v0.2 Performance Snapshot

A short performance snapshot for v0.2. Narrower than a full benchmark dashboard: named local artifacts, graph-vs-Neo4j kept separate from the SurrealDB/pgstack matrix, no single winner number across workloads.

If you want the full benchmark machinery, use:

Snapshots used here

This page is based on two concrete local benchmark snapshots:

Graph vs Neo4j

harness: benchmarks/neo4j-graph-compare/run.py
run id: 20260519-150230
dataset: rows=5000, degree=4, edges=20000
warmup: 2
iterations: 6
artifact: target/benchmarks/neo4j-graph-compare/20260519-150230/report.json

CRUD refresh vs SurrealDB and pgstack

harness: benchmarks/run.sh surreal-suite
run id: 20260519T135910Z
engines: aiondb, surrealdb, pgstack
storage: all three engines on durable local storage
tests: [C]reate, [R]ead, [U]pdate
artifacts:
benchmarks/.state/surreal-suite/20260519T135910Z/summary.tsv
benchmarks/.state/surreal-suite/20260519T135910Z/metadata.json

Broader matrix vs SurrealDB and pgstack

harness: benchmarks/run.sh surreal-suite
run id: full-all-20260512T192959Z
engines: aiondb, surrealdb, pgstack
artifacts:
benchmarks/.state/surreal-suite/full-all-20260512T192959Z/summary.tsv
Benchmark Results

The repository also contains ultra-compare, but this page is based on the two snapshots above because they are the current inspectable artifacts with actual results.

Graph vs Neo4j

The v0.2 graph snapshot favors AionDB on core traversals and graph scans.

All representative queries below kept result parity = true.

Query	AionDB p50	Neo4j p50	Relative result
`out_depth1`	`0.546 ms`	`9.595 ms`	AionDB faster
`out_depth2`	`0.512 ms`	`7.457 ms`	AionDB faster
`out_depth3`	`0.407 ms`	`4.698 ms`	AionDB faster
`in_depth1`	`0.653 ms`	`4.476 ms`	AionDB faster
`edge_filter`	`12.810 ms`	`30.640 ms`	AionDB faster
`multi_out_where`	`15.845 ms`	`34.657 ms`	AionDB faster
`variable_len_4`	`0.549 ms`	`6.024 ms`	AionDB faster
`shortest_path`	`0.927 ms`	`3.022 ms`	AionDB faster

Read this correctly

This is not a claim that AionDB beats Neo4j on every graph workload. It is a narrow claim: on this pinned local graph workload, with this dataset, with this harness, AionDB is ahead on the measured traversal and shortest-path shapes.

Broader matrix vs SurrealDB and pgstack

The broader matrix is mixed, which is why this page keeps it separate.

Fresh CRUD snapshot from surreal-suite:

Workload	AionDB	SurrealDB	pgstack	Read
`[C]reate`	`203.8 ops/s`	`197.0 ops/s`	`486.4 ops/s`	near tie with SurrealDB; pgstack ahead
`[R]ead`	`1409.1 ops/s`	`1840.3 ops/s`	`4545.6 ops/s`	AionDB behind both, but same order of magnitude as SurrealDB
`[U]pdate`	`556.1 ops/s`	`520.1 ops/s`	`1128.1 ops/s`	near tie with SurrealDB; pgstack ahead

Representative broader graph/hybrid results from the full surreal-suite snapshot:

Workload	AionDB	SurrealDB	pgstack	Read
`[S]can::count_all (2000)`	`7819.5 ops/s`	`168.85 ops/s`	`1073.5 ops/s`	AionDB far ahead
`[S]can::graph_edge_filter (2000)`	`1056.55 ops/s`	`219.0 ops/s`	`446.55 ops/s`	AionDB ahead
`[S]can::graph_bidirectional (2000)`	`2874.1 ops/s`	`1322.1 ops/s`	`8.66 ops/s`	AionDB ahead
`[S]can::graph_multi_count (2000)`	`2597.4 ops/s`	`22.82 ops/s`	`532.8 ops/s`	AionDB ahead
`[S]can::graph_multi_out (2000)`	`328.9 ops/s`	`341.9 ops/s`	`UNSUPPORTED`	near tie, slight SurrealDB lead
`[Complex]::graph_two_hop_filter_aggregate (5000)`	`35.20 ops/s`	`UNSUPPORTED`	`0.35 ops/s`	AionDB ahead where supported
`[Complex]::vector_join_graph_filter_rank (5000)`	`61.15 ops/s`	`UNSUPPORTED`	`418.05 ops/s`	pgstack ahead

What this means

The broader v0.2 picture:

graph scans and graph-shaped filters favor AionDB;
CRUD throughput is competitive with SurrealDB on this all-durable refresh; pgstack remains ahead;
hybrid graph/vector is real and runnable, but not a universal win;
some comparison cells are still UNSUPPORTED on one side. Fairness depends on the exact workload family.

Describe v0.2 as:

strong on a growing set of graph and hybrid shapes, mixed as a general benchmark matrix.

Not:

faster than Neo4j, PostgreSQL, and SurrealDB overall.

Short read

against Neo4j on the pinned graph snapshot, AionDB is ahead;
against the broader SurrealDB / pgstack matrix, AionDB wins several graph-heavy scans and loses other workload families;
v0.2 is a graph-evaluation performance story, not a universal cross-engine win.

Reproduce the snapshots

Current Neo4j graph snapshot:

python3 benchmarks/neo4j-graph-compare/run.py \
  --rows 5000 \
  --degree 4 \
  --warmup 2 \
  --iterations 6

Broader SurrealDB / pgstack matrix:

SURREAL_SUITE_ROWS=2000 \
SURREAL_SUITE_WARMUP_SECONDS=3 \
SURREAL_SUITE_ITERATIONS=1 \
SURREAL_SUITE_DURATION_SECONDS=20 \
benchmarks/run.sh surreal-suite

For claim discipline, keep:

the exact command;
the run id;
the commit hash;
the hardware;
the raw output artifacts.