# Experiment Data Analysis



This notebook analyzes the experiment data collected from AWS and IBM cloud environments comparing CAS and Karpenter autoscaling.
import json

import pandas as pd

import matplotlib.pyplot as plt

import seaborn as sns

from pathlib import Path

import numpy as np
# Set up paths

BASE_PATH = Path('.')

AWS_PATH = BASE_PATH / 'aws'

IBM_PATH = BASE_PATH / 'ibm'
def load_json_file(path):

    """Load and parse a JSON file"""

    with open(path) as f:

        return json.load(f)



def parse_latency_data(path):

    """Parse node latency measurement data"""

    data = load_json_file(path)

    return pd.DataFrame(data)



def parse_billing_data(path):

    """Parse AWS billing metrics data"""

    data = load_json_file(path)
    
    # Extract billing data from nested structure
    billing_data = []
    for result in data['ResultsByTime']:
        billing_data.append({
            'start_date': result['TimePeriod']['Start'],
            'end_date': result['TimePeriod']['End'],
            'cost': float(result['Total']['UnblendedCost']['Amount']),
            'currency': result['Total']['UnblendedCost']['Unit'],
            'estimated': result['Estimated']
        })
    
    return pd.DataFrame(billing_data)



def parse_log_file(path):

    """Parse kube-burner log file"""

    with open(path) as f:

        return [line.strip() for line in f if line.strip()]
# Load AWS homogeneous CAS data

aws_homo_cas_path = AWS_PATH / 'homogenous/cas/7e2387d0-1300-43e1-a016-41d8bdba88ae-eks-cas'

aws_cas_latency = parse_latency_data(aws_homo_cas_path / 'nodeLatencyMeasurement-api-intensive.json')

aws_cas_latency_quantiles = parse_latency_data(aws_homo_cas_path / 'nodeLatencyQuantilesMeasurement-api-intensive.json')

aws_cas_billing = parse_billing_data(aws_homo_cas_path / 'aws_billing_metrics_cas.json')
# Load AWS homogeneous Karpenter data

aws_homo_karp_path = AWS_PATH / 'homogenous/cas/7e2387d0-1300-43e1-a016-41d8bdba88ae-eks-karpenter'

aws_karp_latency = parse_latency_data(aws_homo_karp_path / 'nodeLatencyMeasurement-api-intensive.json')

aws_karp_latency_quantiles = parse_latency_data(aws_homo_karp_path / 'nodeLatencyQuantilesMeasurement-api-intensive.json')

aws_karp_billing = parse_billing_data(aws_homo_karp_path / 'aws_billing_metrics_karpenter.json')

aws_karp_logs = parse_log_file(aws_homo_karp_path / 'kube-burner-3c29adda-1a6f-4548-a3a6-0faad0bf8013-iks-karpenter.log')
## Data Analysis



### 1. Node Provisioning Latency Comparison
# Compare latency distributions

plt.figure(figsize=(12, 6))

sns.boxplot(data=[aws_cas_latency['latency'], aws_karp_latency['latency']], 

           labels=['CAS', 'Karpenter'])

plt.title('Node Provisioning Latency Distribution')

plt.ylabel('Latency (seconds)')

plt.show()
### 2. Cost Analysis
# Compare billing metrics

def plot_billing_comparison():

    plt.figure(figsize=(12, 6))

    

    cas_costs = aws_cas_billing['cost'].values

    karp_costs = aws_karp_billing['cost'].values

    

    plt.bar(['CAS', 'Karpenter'], [cas_costs.mean(), karp_costs.mean()])

    plt.title('Average Cost Comparison')

    plt.ylabel('Cost (USD)')

    plt.show()



plot_billing_comparison()
### 3. Performance Quantiles Analysis
# Compare latency quantiles

def plot_quantile_comparison():

    plt.figure(figsize=(12, 6))

    

    quantiles = ['p50', 'p90', 'p95', 'p99']

    cas_values = [aws_cas_latency_quantiles[q].mean() for q in quantiles]

    karp_values = [aws_karp_latency_quantiles[q].mean() for q in quantiles]

    

    x = np.arange(len(quantiles))

    width = 0.35

    

    plt.bar(x - width/2, cas_values, width, label='CAS')

    plt.bar(x + width/2, karp_values, width, label='Karpenter')

    

    plt.xlabel('Percentile')

    plt.ylabel('Latency (seconds)')

    plt.title('Latency Quantiles Comparison')

    plt.xticks(x, quantiles)

    plt.legend()

    plt.show()



plot_quantile_comparison()