Skip to content

What is TurboLynx?

TurboLynx is a fast, scalable OLAP graph database written in C++17.

Key Characteristics

Property Value
Query language Cypher (openCypher subset)
Architecture Single-process embedded (DuckDB-style)
Storage model Extent-based columnar (Graphlet partitioning)
Optimizer ORCA cost-based (Cascades framework, ported from Greenplum)
Graph model Property graph (vertices + directed edges + properties)
Primary use case Read-heavy analytical graph workloads

Architecture at a Glance

Cypher Query
     │
     ▼
  ANTLR4 Parser  →  RegularQuery AST
     │
     ▼
  TurboLynx Binder  →  BoundRegularQuery  (type resolution, schema lookup)
     │
     ▼
  Cypher2OrcaConverter  →  ORCA Logical Plan
     │
     ▼
  ORCA Optimizer  →  Physical Plan  (cost-based join ordering, GEM algorithm)
     │
     ▼
  Pipeline Executor
     │
     ▼
  Storage  (BufferPool → store.db via Linux Kernel AIO)

Graphlets — NULL-free Columnar Storage

Nodes with the same label but different attribute sets are grouped into Graphlets — compact columnar tables with a uniform schema and no NULL columns. The CGC algorithm selects the cost-optimal grouping automatically.

This eliminates the NULL inflation that occurs when storing schemaless graphs in a single wide table, and makes every graphlet SIMD-vectorizable.

GEM — Per-Graphlet Join Ordering

The GEM (Graph-aware Enumeration via Memo) optimizer extension pushes joins below UNION ALL nodes, allowing each graphlet group to receive its own cost-optimal join order. This is particularly effective on schemaless graphs where different graphlets have very different cardinalities.

Single-Process Embedded

Unlike traditional graph databases that run as a separate server process, TurboLynx embeds directly into the calling process — similar to DuckDB or SQLite.

  • No daemon to start or stop
  • No IPC or shared memory
  • Catalog persisted as catalog.bin; graph data as store.db
  • C API via libturbolynx.so

No External Runtime Dependencies

All dependencies are compiled into the binary from bundled source. Only standard C/C++ runtime libraries (libstdc++, libc, libpthread) are required at runtime.

Next Steps