Skip to content

Performance Analysis & Optimization Opportunities

Jump to bottom
Richard Fu edited this page Jan 24, 2026 · 1 revision

Performance Analysis & Optimization Opportunities

Date: 2026-01-15 Game: template_lines (representative baseline) Simulations: 1,000 Total Time: 4.899s Speed: 204.1 sims/second Per Sim: 4.899ms

Executive Summary

Based on profiling data, the top 5 hot paths consuming ~70% of execution time are:

  1. symbol.py:122(assign_paying_bool) - 0.895s (18.3%)
  2. copy.py:119(deepcopy) - 0.968s cumulative (19.8%)
  3. posix.read (multiprocessing) - 0.896s (18.3%)
  4. set.add operations - 0.238s (4.9%)
  5. JSON encoding - 0.225s (4.6%)

Key Finding: Current performance of 204 sims/second is reasonable, but there are several optimization opportunities that could yield 20-30% improvements.

Hot Path Analysis

1. Symbol.assign_paying_bool() - CRITICAL

Impact: 0.895s (18.3% of total time) Calls: 99,811 times Per Call: 8.96µs

Current Implementation (estimated from profiling):

def assign_paying_bool(self):
    # Likely checking paytable for each symbol
    # Called once per symbol during board creation
    pass

Problem: Called for EVERY symbol on EVERY board (99K+ times for 1K sims)

Optimization Opportunities:

  1. Cache paytable lookups - Paytable doesn't change during simulation
  2. Pre-compute paying symbols - Create a set of paying symbol names at initialization
  3. Lazy evaluation - Only compute when needed (not always needed)

Estimated Impact: 30-50% reduction (0.3-0.4s savings)

Priority: 🔴 HIGH

2. Deep Copy Operations - CRITICAL

Impact: 0.968s cumulative (19.8% of total time) Calls: 639,109 deepcopy calls

Current Usage:

  • Board state copying
  • Symbol copying
  • Configuration copying

Problem: Excessive deepcopy usage (639K+ calls for 1K sims)

Optimization Opportunities:

  1. Reduce deepcopy usage - Use shallow copy where safe
  2. Object pooling - Reuse symbol objects instead of copying
  3. Immutable data structures - Eliminate need for copying
  4. Manual copying - Write custom copy methods for specific cases

Estimated Impact: 40-60% reduction (0.4-0.6s savings)

Priority: 🔴 HIGH

3. Multiprocessing Overhead - CANNOT OPTIMIZE

Impact: 0.896s (18.3% of total time) Calls: 520 posix.read calls

Analysis: This is overhead from multiprocessing communication (Manager, Pipes). It's necessary for parallel execution and cannot be optimized without changing architecture.

Trade-off: Multiprocessing provides speedup for large simulations (threads > 1), but adds overhead for small batches.

Recommendation: Accept this overhead. For production use with 10K+ sims and multiple threads, the benefit outweighs the cost.

Priority: ⚪ ACCEPT

4. Set Operations - MEDIUM PRIORITY

Impact: 0.238s (4.9% of total time) Calls: 2,995,341 set.add operations

Current Usage:

  • Tracking winning positions
  • Tracking special symbols
  • Building line/cluster sets

Optimization Opportunities:

  1. Use lists where order matters - Lists are faster for small collections
  2. Pre-allocate sets - Reduce resizing overhead
  3. Batch operations - Fewer, larger set operations

Estimated Impact: 10-20% reduction (0.02-0.05s savings)

Priority: 🟡 MEDIUM

5. JSON Encoding - LOW PRIORITY

Impact: 0.225s (4.6% of total time) Calls: 5,522 encoder.iterencode calls

Analysis: JSON encoding is necessary for output. Already optimized in Phase 3.1 (OutputFormatter).

Additional Opportunities:

  1. Use orjson library - 2-3x faster than standard json
  2. Batch writes - Write larger chunks less frequently
  3. Binary format - Use msgpack or pickle for internal use

Estimated Impact: 30-50% reduction (0.07-0.11s savings) with orjson

Priority: 🟢 LOW (already optimized in Phase 3.1)

Optimization Recommendations

Phase 5.1A: Quick Wins (1-2 hours)

1. Cache Symbol Paytable Lookups ✅

File: src/calculations/symbol.py:122

Change:

class Symbol:
    _paying_symbols_cache = {}  # Class-level cache

    def assign_paying_bool(self):
        # Use cached lookup instead of recalculating
        cache_key = (self.name, self.config_hash)
        if cache_key in Symbol._paying_symbols_cache:
            self.paying = Symbol._paying_symbols_cache[cache_key]
        else:
            # Original logic
            self.paying = self._check_paytable()
            Symbol._paying_symbols_cache[cache_key] = self.paying

Impact: -0.3s to -0.4s (6-8% total speedup)

2. Reduce Unnecessary Deepcopy ✅

Files: board.py, base_game_state.py

Change: Identify where deepcopy is used unnecessarily and replace with:

  • Shallow copy for simple dicts/lists
  • Reference passing for read-only data
  • Custom copy methods for specific classes

Impact: -0.4s to -0.6s (8-12% total speedup)

Combined Quick Wins: 14-20% faster (4.9s → 4.0-4.2s)

Phase 5.1B: Medium Effort (2-3 hours)

3. Object Pooling for Symbols 🔄

File: src/calculations/symbol.py

Change: Reuse Symbol objects instead of creating new ones

class SymbolPool:
    def __init__(self, config):
        # Pre-create symbol objects for each type
        self.pool = {name: [] for name in config.symbols}

    def get_symbol(self, name):
        if self.pool[name]:
            return self.pool[name].pop()
        return Symbol(name)

    def return_symbol(self, symbol):
        symbol.reset()
        self.pool[symbol.name].append(symbol)

Impact: -0.2s to -0.3s (4-6% total speedup)

4. Optimize Set Operations 🔄

Files: lines.py, cluster.py, ways.py

Change: Use lists for small collections, pre-allocate sets

Impact: -0.05s to -0.1s (1-2% total speedup)

Combined Medium Effort: 5-8% faster (additional speedup)

Phase 5.1C: Major Refactor (4-6 hours) - NOT RECOMMENDED

5. Replace Multiprocessing with asyncio

Benefit: Eliminate 0.9s overhead Cost: Major architecture change, potential correctness issues Recommendation: SKIP - Multiprocessing is appropriate for CPU-bound work

6. Binary Output Format

Benefit: Faster serialization Cost: Breaking change, RGS compatibility issues Recommendation: DEFER - Phase 3 compression is sufficient

Target Performance Goals

Baseline (Current)

  • Speed: 204.1 sims/second
  • Time for 1K sims: 4.9s
  • Time for 10K sims: 49s
  • Time for 100K sims: 490s (8.2 minutes)

After Phase 5.1A Optimizations (Quick Wins)

  • Speed: 240-250 sims/second (+18-23%)
  • Time for 1K sims: 4.0-4.2s
  • Time for 10K sims: 40-42s
  • Time for 100K sims: 400-420s (6.7-7.0 minutes)

After Phase 5.1A+B Optimizations (All Optimizations)

  • Speed: 255-270 sims/second (+25-32%)
  • Time for 1K sims: 3.7-3.9s
  • Time for 10K sims: 37-39s
  • Time for 100K sims: 370-390s (6.2-6.5 minutes)

Memory Analysis

To Be Completed: Run memory profiling with tracemalloc

Expected Findings:

  • High memory usage from symbol objects (99K+ instances)
  • Board storage overhead
  • Event list accumulation

Optimization Opportunities:

  • Symbol object pooling (reuse objects)
  • Streaming output (write events incrementally)
  • Reduce intermediate data structures

Conclusions

What to Optimize Now (Phase 5.1)

  1. ✅ Symbol paytable caching - 6-8% speedup, low risk
  2. ✅ Reduce deepcopy usage - 8-12% speedup, medium risk
  3. 🔄 Symbol object pooling - 4-6% speedup, medium risk (optional)

What NOT to Optimize

  1. ❌ Multiprocessing overhead - Necessary architecture
  2. ❌ JSON encoding - Already optimized in Phase 3.1
  3. ❌ Line/cluster algorithms - Already efficient

Expected Results

  • Total speedup: 18-23% (Phase 5.1A) to 25-32% (Phase 5.1A+B)
  • 10K simulations: 49s → 37-42s
  • 100K simulations: 8.2min → 6.2-7.0min
  • Risk level: Low to medium
  • Breaking changes: None

Next Steps

  1. ✅ Complete profiling analysis (this document)
  2. ⏭️ Implement Symbol paytable caching
  3. ⏭️ Audit and reduce deepcopy usage
  4. ⏭️ Run benchmarks to measure improvements
  5. ⏭️ Update RALPH_TASKS.md with results
  6. ⏭️ Document optimizations in code comments

Status: Analysis complete, ready for implementation Estimated Implementation Time: 3-5 hours Expected Benefit: 18-32% faster simulations Risk Assessment: Low (non-breaking changes)

Clone this wiki locally