Test 3: GPU-Accelerated PCAP Scanner

Overview

This folder contains the successful GPU-accelerated PCAP scanner implementation that achieved 2,741 MB/s peak throughput - a 45x improvement over CPU implementation. The scanner uses CuPy (CUDA) for GPU acceleration with adaptive algorithms.

Files Explained

newtest.py - Main GPU Scanner

• Purpose: The actual GPU-accelerated PCAP scanner implementation
• Technology: Uses CuPy (CUDA) for GPU acceleration
• Algorithms:
  • ≤16 patterns: Boyer-Moore-Horspool on GPU (one pass per pattern)
  • ≥17 patterns: PFAC (Parallel Finite Automaton for Content) on GPU
• Features:
  • Loads PCAP files using memory mapping
  • Processes patterns on GPU using CUDA kernels
  • Outputs results to CSV with timing and throughput metrics
  • Handles both small and large packet files efficiently

run_comprehensive_test.py - Test Automation Script

• Purpose: Automates running newtest.py across all test combinations
• Function:
  • Defines 3 pattern sets (1, 7, 14 patterns)
  • Defines 10 PCAP files (5 small packet + 5 large packet)
  • Runs 30 total test combinations (10 files × 3 pattern counts)
  • Executes python newtest.py commands with different parameters
  • Saves results to timestamped CSV file
• Output: Generates gpu_test_results_2025-09-12_12-47-35.csv

gpu_test_results_2025-09-12_12-47-35.csv - Test Results Data

• Purpose: Contains the actual performance results from all 30 test runs
• Columns:
  • pcap_file, file_size_mb, num_patterns
  • load_time, search_time, total_time
  • throughput_mbps, num_matches, num_packets
• Data: Shows the 2,741 MB/s peak performance and all other results
• Usage: Raw data for analysis and reporting

GPU_PCAP_Performance_Report_2025-09-12.md - Analysis Report

• Purpose: Human-readable analysis of the test results
• Content:
  • Executive summary with key findings
  • Performance analysis by file type and pattern count
  • Technical insights and recommendations
  • Detailed breakdown of results
• Value: Explains what the numbers mean and why performance varies

How They Work Together

1. run_comprehensive_test.py orchestrates the testing
2. newtest.py does the actual GPU-accelerated scanning
3. gpu_test_results_*.csv stores the raw performance data
4. GPU_PCAP_Performance_Report_*.md provides human-readable analysis

Key Achievements

• Peak throughput: 2,741 MB/s (500MB large packets, single pattern)
• 45x improvement over CPU implementation (61 MB/s)
• Scalability: Handles files up to 1GB efficiently
• Adaptive algorithms: BMH for ≤7 patterns, PFAC for ≥14 patterns
• Match accuracy: Over 21 million matches found across all tests

Usage

# Run comprehensive test
python run_comprehensive_test.py

# Run single test
python newtest.py "PCAP Files/synthetic_200mb.pcapng" -s password --csv-output results.csv

Requirements

• Python 3.9+
• CuPy (CUDA 13.x)
• NVIDIA GPU with CUDA support
• PCAP/PCAPNG files for testing

This is a complete testing and analysis pipeline that demonstrates the GPU scanner’s capabilities across different workloads and provides both raw data and interpreted results.