rusty_feeder::optimization

Module zero_alloc_parser

Source
Expand description

Zero-allocation JSON parser for market data messages Zero-allocation WebSocket message parser for HFT applications

This module provides ultra-fast JSON parsing optimized for high-frequency trading where allocation overhead can destroy microsecond-level latency requirements.

§HFT Performance Rationale

§Latency Impact of Memory Allocation

In HFT systems, every nanosecond counts:

  • Heap allocation: 50-200ns overhead per allocation
  • Garbage collection: Unpredictable 1-10ms pauses
  • Memory fragmentation: Increased cache misses and TLB pressure
  • Lock contention: Global allocator locks block trading threads

§Critical Path Optimization

Market data processing must complete within:

  • Trade messages: <500ns from receipt to strategy signal
  • Order book updates: <200ns for L2 depth changes
  • Ticker updates: <100ns for price/volume notifications

§Zero-Allocation Architecture

§Buffer Pool Management

  • Pre-allocated buffers: Eliminates malloc/free in hot paths
  • Buffer recycling: LIFO stack for optimal cache locality
  • Size optimization: Configurable buffer sizes for different message types
  • Growth prevention: Caps maximum buffer size to prevent memory bloat

§SIMD-JSON Integration

  • simd_json library: 2-10x faster than serde_json
  • SIMD parsing: Vectorized JSON tokenization and validation
  • Borrowed values: Zero-copy string/number extraction
  • In-place parsing: Modifies input buffer directly (no extra allocation)

§Lock-Free Data Structures

  • DashMap type cache: Lock-free concurrent HashMap for message type recognition
  • Atomic statistics: Lock-free performance monitoring
  • Thread-local pools: Per-thread buffer pools eliminate contention

§Performance Characteristics

§Typical Latency Savings

  • 40-60% reduction in JSON parsing time vs. traditional approaches
  • 90%+ cache hit rate for message type recognition
  • Zero allocation in steady-state operation
  • Sub-100ns parsing for typical market data messages

§Memory Efficiency

  • Predictable memory usage: Pre-allocated pool sizes
  • Cache-friendly access: Buffer reuse improves L1/L2 cache hit rates
  • Reduced memory bandwidth: Eliminates allocation/deallocation traffic

§Threading Model

§Thread-Local Optimization

  • Per-thread parsers: Eliminates lock contention
  • CPU cache affinity: Buffers stay warm in thread-local cache
  • Lock-free operation: No synchronization overhead

§Concurrent Safety

  • Send + Sync: Safe sharing across threads when needed
  • Arc wrapping: Shared parser instances for specific use cases
  • Atomic counters: Lock-free statistics aggregation

§Exchange Integration

Optimized for common exchange message patterns:

  • Coinbase Pro: Match, L2Update, Heartbeat messages
  • Binance: Trade, Depth, Ticker streams
  • Bybit: Trade, OrderBook, Kline updates
  • Generic protocols: Extensible message type recognition

§Monitoring & Observability

Built-in performance metrics:

  • Parse latency: Nanosecond-precision timing
  • Cache hit rates: Type recognition efficiency
  • Buffer reuse: Memory allocation avoidance
  • Zero-copy operations: Borrowed value usage tracking

Structs§

AtomicParserStats
Lock-free atomic statistics for parser performance
ParserStats
Parser performance statistics
ZeroAllocParser
Zero-allocation message parser with lock-free data structures

Enums§

MessageType
Pre-allocated message types for zero-copy parsing

Functions§

extract_trade_fields
Extract common fields from message using thread-local parser
parse_message_fast
Parse message using thread-local parser (highest performance)
parse_message_fast_with_closure
Parse message using thread-local parser with closure for zero-copy processing
parser_stats
Get thread-local parser statistics