Performance

Benchmarked with BenchmarkDotNet on Apple M1 Pro, .NET 10.0.5, Release build.

Results

Operation	Mean	Allocated	Notes
list_graphs	62 ns	216 B	TTL cached (10s)
get_schema	117 ns	344 B	TTL cached (10s)
get_or_create_graph (exists)	775 us	5.6 KB	SQL check
cypher_query (1 vertex)	1.0 ms	58 KB	Single MATCH + RETURN
search_vertices (limit=2)	1.0 ms	36 KB
search_vertices (all, 16)	1.3 ms	194 KB
search_edges (all, 22)	1.6 ms	153 KB
get_neighbors (depth=1)	92 ms	233 KB	Directed UNION + batch connection
export_graph (40 entities)	71 ms	472 KB	Batch connection

Optimizations Applied

TTL Cache (list_graphs, get_schema)

Metadata queries are cached for 10 seconds. Cache is automatically invalidated on write operations (upsert_vertex, drop_graphs, etc.). This reduces repeated calls from ~1ms to ~100ns.

Connection Batching (get_neighbors, export_graph, generate_visualization)

Operations that run multiple Cypher queries reuse a single database connection. This eliminates the per-query overhead of LOAD 'age' + SET search_path (~0.5ms per connection).

Directed Edge Queries (get_neighbors)

Bidirectional edge patterns (-[e]-) are slow in Apache AGE. get_neighbors uses directed UNION ALL queries instead:

-- Before: ~260ms (bidirectional scan)
MATCH (start {ident: 'x'})-[e]-(neighbor) RETURN e

-- After: ~92ms (directed UNION)
MATCH (start {ident: 'x'})-[e]->(neighbor) RETURN e
UNION ALL
MATCH (start {ident: 'x'})<-[e]-(neighbor) RETURN e

ReadyToRun

The tool is published with PublishReadyToRun=true for faster cold start via pre-JIT compilation.

Running Benchmarks

cd bench
AGE_CONNECTION_STRING="Host=localhost;Port=5435;Database=agemcp;Username=agemcp;Password=agemcp" \
  dotnet run -c Release

Results are written to bench/BenchmarkDotNet.Artifacts/.