Skip to content

performance.md

Latest commit

 

History

History
249 lines (203 loc) · 10.8 KB

performance.md

File metadata and controls

249 lines (203 loc) · 10.8 KB
title description
Performance benchmarks
The Stock Indicators for .NET library is built for speed and production workloads. Compare our execution times with other options.

Performance benchmarks for v3.0.0

These are the execution times for the current indicators using two years of historical daily stock quotes (502 periods) with default or typical parameters.

Series Style Benchmarks (v2/v3)

The following benchmarks represent traditional batch processing performance for complete historical datasets:

BenchmarkDotNet v0.14.0, Ubuntu 22.04.5 LTS (Jammy Jellyfish)
AMD EPYC 7763, 1 CPU, 4 logical and 2 physical cores
.NET SDK 9.0.100
  [Host]   : .NET 9.0.0 (9.0.24.52809), X64 RyuJIT AVX2
Method Mean Error StdDev
GetAdl 34.89 μs 2.409 μs 0.132 μs
GetAdx 42.48 μs 2.417 μs 0.132 μs
GetAlligator 32.84 μs 0.380 μs 0.021 μs
GetAlma 27.08 μs 0.524 μs 0.029 μs
GetAroon 54.64 μs 0.722 μs 0.040 μs
GetAtr 36.40 μs 4.916 μs 0.269 μs
GetAtrStop 52.84 μs 5.843 μs 0.320 μs
GetAwesome 48.42 μs 0.809 μs 0.044 μs
GetBeta 164.01 μs 23.671 μs 1.297 μs
GetBetaUp 128.36 μs 8.187 μs 0.449 μs
GetBetaDown 133.49 μs 5.559 μs 0.305 μs
GetBetaAll 357.86 μs 10.892 μs 0.597 μs
GetBollingerBands 69.14 μs 2.992 μs 0.164 μs
GetBop 37.33 μs 3.893 μs 0.213 μs
GetCci 46.97 μs 3.619 μs 0.198 μs
GetChaikinOsc 55.48 μs 2.268 μs 0.124 μs
GetChandelier 53.86 μs 6.396 μs 0.351 μs
GetChop 59.92 μs 0.953 μs 0.052 μs
GetCmf 70.13 μs 7.516 μs 0.412 μs
GetCmo 33.04 μs 2.522 μs 0.138 μs
GetConnorsRsi 109.98 μs 2.155 μs 0.118 μs
GetCorrelation 96.95 μs 7.741 μs 0.424 μs
GetDema 22.99 μs 1.052 μs 0.058 μs
GetDoji 69.18 μs 5.349 μs 0.293 μs
GetDonchian 204.30 μs 10.312 μs 0.565 μs
GetDpo 48.21 μs 1.477 μs 0.081 μs
GetDynamic 39.30 μs 1.816 μs 0.100 μs
GetElderRay 53.86 μs 4.422 μs 0.242 μs
GetEma 22.31 μs 1.314 μs 0.072 μs
GetEpma 55.52 μs 3.782 μs 0.207 μs
GetFcb 217.82 μs 6.294 μs 0.345 μs
GetFisherTransform 52.84 μs 1.077 μs 0.059 μs
GetForceIndex 32.86 μs 2.649 μs 0.145 μs
GetFractal 48.06 μs 5.058 μs 0.277 μs
GetGator 59.64 μs 9.488 μs 0.520 μs
GetHeikinAshi 109.37 μs 1.923 μs 0.105 μs
GetHma 99.47 μs 5.452 μs 0.299 μs
GetHtTrendline 84.69 μs 18.501 μs 1.014 μs
GetHurst 838.65 μs 127.955 μs 7.014 μs
GetIchimoku 640.07 μs 44.858 μs 2.459 μs
GetKama 30.06 μs 4.795 μs 0.263 μs
GetKlinger 43.21 μs 2.307 μs 0.126 μs
GetKeltner 64.88 μs 21.930 μs 1.202 μs
GetKvo 43.35 μs 3.056 μs 0.168 μs
GetMacd 46.94 μs 4.013 μs 0.220 μs
GetMaEnvelopes 43.99 μs 1.648 μs 0.090 μs
GetMama 75.49 μs 1.308 μs 0.072 μs
GetMarubozu 78.81 μs 3.513 μs 0.193 μs
GetMfi 43.31 μs 1.323 μs 0.073 μs
GetObv 33.07 μs 1.152 μs 0.063 μs
GetObvWithSma 42.54 μs 1.500 μs 0.082 μs
GetParabolicSar 36.18 μs 0.569 μs 0.031 μs
GetPivotPoints 45.84 μs 2.456 μs 0.135 μs
GetPivots 87.39 μs 1.025 μs 0.056 μs
GetPmo 39.11 μs 1.240 μs 0.068 μs
GetPrs 40.60 μs 2.618 μs 0.144 μs
GetPrsWithSma 43.07 μs 1.737 μs 0.095 μs
GetPvo 44.81 μs 2.604 μs 0.143 μs
GetRenko 59.82 μs 3.289 μs 0.180 μs
GetRenkoAtr 67.58 μs 24.816 μs 1.360 μs
GetRoc 24.04 μs 1.218 μs 0.067 μs
GetRocWb 40.13 μs 3.526 μs 0.193 μs
GetRocWithSma 30.52 μs 2.540 μs 0.139 μs
GetRollingPivots 243.31 μs 37.635 μs 2.063 μs
GetRsi 25.91 μs 0.812 μs 0.044 μs
GetSlope 56.35 μs 9.701 μs 0.532 μs
GetSma 24.90 μs 2.631 μs 0.144 μs
GetSmaAnalysis 42.43 μs 0.780 μs 0.043 μs
GetSmi 37.50 μs 1.633 μs 0.089 μs
GetSmma 23.26 μs 2.726 μs 0.149 μs
GetStarcBands 71.22 μs 27.014 μs 1.481 μs
GetStc 82.32 μs 4.596 μs 0.252 μs
GetStdDev 71.29 μs 1.644 μs 0.090 μs
GetStdDevWithSma 76.73 μs 1.728 μs 0.095 μs
GetStdDevChannels 66.38 μs 8.307 μs 0.455 μs
GetStoch 51.89 μs 3.650 μs 0.200 μs
GetStochSMMA 50.08 μs 3.252 μs 0.178 μs
GetStochRsi 61.82 μs 3.847 μs 0.211 μs
GetSuperTrend 54.21 μs 11.806 μs 0.647 μs
GetT3 26.51 μs 0.176 μs 0.010 μs
GetTema 23.74 μs 1.415 μs 0.078 μs
GetTr 33.63 μs 5.232 μs 0.287 μs
GetTrix 26.56 μs 1.217 μs 0.067 μs
GetTrixWithSma 30.39 μs 1.462 μs 0.080 μs
GetTsi 29.67 μs 6.213 μs 0.341 μs
GetUlcerIndex 95.21 μs 2.543 μs 0.139 μs
GetUltimate 58.52 μs 2.435 μs 0.133 μs
GetVolatilityStop 61.17 μs 22.487 μs 1.233 μs
GetVortex 43.12 μs 1.817 μs 0.100 μs
GetVwap 33.56 μs 1.656 μs 0.091 μs
GetVwma 42.38 μs 1.098 μs 0.060 μs
GetWilliamsR 52.27 μs 0.135 μs 0.007 μs
GetWma 31.98 μs 2.411 μs 0.132 μs
GetZigZag 99.35 μs 5.101 μs 0.280 μs

Streaming Performance Characteristics (v3)

v3 introduces BufferList and StreamHub styles for incremental and real-time processing. Here's how they compare:

Performance Comparison

Style Use Case Relative Performance Latency per Quote
Series Batch processing Baseline (fastest) N/A
BufferList Incremental updates ~10-20% overhead <100μs typical
StreamHub Real-time coordination ~20-30% overhead <1ms typical

BufferList vs Series

BufferList style provides efficient incremental processing with modest overhead:

Advantages:

  • O(1) or O(log n) per-quote updates for most indicators
  • Automatic buffer management and memory pruning
  • No need to recalculate entire history on each update
  • Memory-efficient for growing datasets

Performance Profile:

Series (502 quotes):      ~25μs total
BufferList (502 quotes):  ~30μs total (~20% overhead)
Per-quote latency:        ~60ns average

Example indicators:

  • SMA: O(1) updates with rolling sum
  • EMA: O(1) updates with weighted smoothing
  • RSI: O(1) updates with Wilder's smoothing
  • Bollinger Bands: O(1) updates with rolling statistics

StreamHub vs Series

StreamHub style adds observable patterns and state management:

Advantages:

  • Single quote update propagates to multiple observers
  • Built-in rollback support for late-arriving data
  • Indicator chaining with automatic updates
  • Optimized for low-latency real-time scenarios

Performance Profile:

Series (502 quotes):      ~25μs total
StreamHub (502 quotes):   ~32μs total (~28% overhead)
Per-quote latency:        ~64ns average
Rollback (late Add):      ~2-5μs for state rebuild

Scaling Characteristics:

  • Hub overhead is amortized across multiple observers
  • 5 indicators on 1 hub ≈ 1.5× single Series calculation
  • Linear scaling with number of quotes
  • Cache size grows with lookback periods

Memory Overhead

Typical memory footprint per indicator instance:

Style Memory per Instance Scaling Factor
Series ~4KB (results only) N/A
BufferList ~8KB (buffers + results) Grows with lookback
StreamHub ~12KB (cache + state + results) Grows with lookback

Memory optimization tips:

  • Set MaxListSize on BufferList to limit result history
  • Use .Clear() periodically to reset state
  • Consider Series style for one-time historical analysis
  • Chain indicators to avoid duplicate calculations

Latency Targets

Real-time performance targets for trading applications:

Scenario Target Typical Performance
Single indicator per quote <100μs 60-80μs
5 indicators on hub per quote <500μs 300-400μs
Complex chains (EMA→RSI→Slope) <200μs 120-150μs
State rebuild (Add/Remove) <5ms 2-3ms

When to Choose Each Style

Choose Series when:

  • Processing complete historical datasets
  • Backtesting with no real-time requirements
  • One-time calculations or periodic batch updates
  • Maximum throughput is priority

Choose BufferList when:

  • Building up data incrementally
  • Single indicator with growing dataset
  • Memory efficiency is important
  • Simple incremental updates without coordination

Choose StreamHub when:

  • Multiple indicators need synchronized updates
  • Live data feeds or WebSocket integration
  • Low latency per quote is critical
  • State management and rollback support needed

Benchmarking Notes

All benchmarks performed on:

  • AMD EPYC 7763, 1 CPU, 4 logical and 2 physical cores
  • .NET 9.0.0 (9.0.24.52809), X64 RyuJIT AVX2
  • Ubuntu 22.04.5 LTS (Jammy Jellyfish)
  • 502 periods of historical daily quotes
  • Default or typical indicator parameters

Performance may vary based on:

  • Indicator complexity and lookback periods
  • Quote frequency (tick, minute, hour, daily)
  • Hardware specifications and .NET version
  • Number of concurrent calculations