A lean, zero-dependency C database core built to eliminate database bloat. Engineered to handle high-concurrency ingestion and vector workloads seamlessly across Windows, Linux, Android, and macOS/iOS—bridged cleanly via FFI into Python, PHP, C#, Node.js, Rust, and Flutter.
Decoupled by design. Built to do one thing with absolute precision rather than pretending to be an all-in-one monolith.
No bloated monolithic binaries. Physical storage and indexing are completely decoupled from network protocols and query orchestration. Interfaced seamlessly via lightweight Adapters and C-ABI FFI.
Ending corruption nightmares. Strict 64-byte L1 cache-line and 4K physical block alignment eliminate write amplification, backed by inline 64-bit FNV-1a row-level fingerprints and automated shadow mounting.
Engineered for raw throughput. Utilizes dynamic Slot Ring buffers and user-space atomic operations, directly dispatching physical writes to Linux io_uring and Windows IOCP with zero mutex lock contention.
Physical ingestion tested on a 15W ultrabook (Intel i7-8650U, 4C/8T) using 46,173 real Taiwan stock market records under concurrent multi-threaded write contention.
| Language / Runtime | Interface Driver | Latency (ms) | Throughput (TPS) | Practical Takeaway |
|---|---|---|---|---|
| C Core (Native) | Direct C-ABI / Slot Ring | 1,031.46 ms | 44,764.50 | Pure bare-metal hardware saturation, zero heap allocation |
| Linux VM (io_uring) | Pre-allocated Ring Slot | ~1,180.00 ms | 38,000.00+ | Shared submission queues bypass hypervisor syscall stalls |
| C++ OO | Object Class Wrapper | 1,259.43 ms | 36,661.79 | Zero-copy object encapsulation |
| Flutter / Dart | Dart FFI Isolate | 1,612.16 ms | 28,640.42 | Direct isolate dispatch for GUI apps without dropping UI frames |
| Python FFI | ctypes Native Wrapper | 2,138.12 ms | 21,595.05 | Atomic reservation bypassing Python GIL contention |
| PHP (Web UI / FFI) | Direct C Heap Allocation | 3,815.17 ms | 12,102 ~ 26,000+ | In-process execution with zero MySQL connection pool overhead |
The best way to prove a core engine is to build an entire multi-platform tooling suite on top of it.
ez_Shell), Native Desktop/Mobile GUIs (Windows, Linux, macOS, Android, iOS), and the Web (dbu_web.php). It provides visual schema maintenance, microsecond EDX tag seeking, and declarative Relation Studio views compiled into compact .ERL binaries.