GPU vs CPU Performance Analysis - January 5, 2025, 15:30

This folder contains a complete analysis of GPU vs CPU performance for PCAP pattern matching, including all code and data used to generate the results.

Quick Reproduction Steps

1. Install Dependencies:

   pip install cupy-cuda12x scapy pandas colorama

2. Run Complete Benchmark:

   python comprehensive_pattern_benchmark.py

3. Run Specific Test:

   python comprehensive_pattern_benchmark.py --test-id test_001

4. Resume Interrupted Benchmark:

   python comprehensive_pattern_benchmark.py --resume-from test_019

5. Quick Validation (first 3 tests):

   python comprehensive_pattern_benchmark.py --quick

Note: Requires NVIDIA GPU with CUDA support. Results saved to comprehensive_benchmark_progress.csv.

Contents

Analysis Report

• GPU_vs_CPU_Performance_Analysis_2025-01-05_15-30.md - Complete analysis report with findings, methodology, and conclusions

Benchmark Code

• comprehensive_pattern_benchmark.py - Main benchmark script that orchestrates CPU vs GPU comparisons
• pcap_gpu_scanner.py - GPU-accelerated PCAP scanner implementation
• gpu_kernels.py - CUDA kernel implementations for pattern matching algorithms

PCAP Generation Scripts

• create_large_synthetic_pcap.py - Generates PCAP files with small packets (43-139 bytes average)
• create_large_packet_pcap.py - Generates PCAP files with large packets (~26KB average)

Data Files

• comprehensive_benchmark_progress.csv - Complete benchmark results from 25 test scenarios

Key Findings

• Packet Size Critical: GPU performance heavily depends on packet characteristics
• Small Packets: GPU is 20-300x slower than CPU and consistently times out
• Large Packets: GPU can be competitive (0.19-1.11x speedup) on smaller files
• Algorithm Selection: Dynamic algorithm selection based on pattern count
• Validation Issues: 40% of tests failed due to GPU timeouts

Test Coverage

• 25/30 tests completed (83%)
• PCAP Sizes: 50MB, 100MB, 200MB, 500MB, 1000MB
• Packet Types: Small packets (43-139 bytes) vs Large packets (~26KB)
• Pattern Counts: 1, 7, 14 patterns
• Algorithms: Boyer-Moore-Horspool, Aho-Corasick, Vectorized Multi-Pattern, PFAC

Reproducibility

All code is included to reproduce the analysis. See the methodology section in the main report for detailed execution instructions.

Generated

• Date: January 5, 2025
• Time: 15:30
• Status: Analysis complete, 5 tests remaining (1GB large packet scenarios)