Maph - Memory-Mapped Approximate Perfect Hash

Maph is a high-performance, memory-mapped database and framework for space-efficient approximate data structures with O(1) lookups. Built on perfect hash functions, it provides sub-microsecond access to arbitrary mappings (X→Y) with configurable storage and accuracy trade-offs.

Core Features

• O(1) Lookups: Guaranteed constant-time operations with perfect hash optimization
• Memory-Mapped Storage: Zero-copy access via mmap for minimal memory overhead
• Lock-Free Operations: Atomic operations for thread-safe concurrent access
• Configurable Storage: 8/16/32/64-bit options for space/accuracy trade-offs
• JSON Database: Built-in daemon with REST API for key-value storage

Performance

• GET: 10M ops/sec (<100ns latency)
• SET: 8M ops/sec (<150ns latency)
• Batch operations: 50M ops/sec with SIMD optimization

Usage Example

#include <maph.hpp>

// Create a new maph store
auto store = maph::Maph::create("mystore.maph", 1000000);

// Store JSON key-value pairs
store->set(R"({"user": "alice"})", R"({"score": 95})");

// Retrieve with O(1) lookup
auto value = store->get(R"({"user": "alice"})");

// Batch operations for high throughput
std::vector<std::pair<JsonView, JsonView>> batch = {
    {R"({"id": 1})", R"({"name": "item1"})"},
    {R"({"id": 2})", R"({"name": "item2"})"}
};
store->mset(batch);
Copy

Applications

• High-performance caching with sub-microsecond lookups
• ML feature stores with guaranteed latency
• Bloom filter alternative with configurable accuracy
• Rate limiting with O(1) token bucket checks

View on GitHub | Documentation