AWS EMR scale benchmarks (50GB+)

[AayushBarhate]

What

Run all approaches on AWS EMR (m5.xlarge x3) with 50GB+ data. 3 traffic scenarios × 4 approaches × 5 runs = 60 benchmark runs.

Why

Local testing on 2-core VM doesn't prove the system works at production scale. Paper needs scale numbers.

Steps

1. Provision EMR cluster (3x m5.xlarge) + MSK Kafka
2. Deploy agent JAR, predictor, generator
3. Run experiment matrix: 3 scenarios (steady, periodic burst, random burst) × 4 approaches × 5 runs
4. Collect latency CDFs, throughput curves, CPU/memory profiles
5. Run Wilcoxon signed-rank tests for statistical significance

Acceptance criteria

• 60 completed runs with raw data
• Publication-quality latency CDF plots
• Statistical significance confirmed (p < 0.05)