Function getCurve

artifact_evaluation/plot_azure_500.py:50–86

(path, idx=0)

def getCurve(path, idx=0):
    df = pd.read_csv(path)

    # discard first 10 minutes
    df = df[~df.invocationID.str.contains("^min[0-9]\.", regex=True)]

    print(path)
    print(f"Failure percentage: {df[df['connectionTimeout'] | df['functionTimeout']].shape[0] / df.shape[0]}")

    df['function_hash'] = df['instance'].str.split("-").str[idx]
    df = df.reset_index(drop=True)
    df['slowdown'] = df['responseTime'] / df['requestedDuration']
    print(f"Slowdown < 1: {df[df['slowdown'] < 1].shape[0] / df.shape[0]}")

    # number of instances created
    print(f"Number of instances created: {df['instance'].nunique()}")

    # filter functions with invalid slowdown
    df = df[df['slowdown'] >= 1]

    df = df.groupby(df.function_hash).slowdown.apply(stats.gmean)
    df = df.to_frame()

    # filter warm starts
    # beforeWarmFilter = df.shape[0]
    # df = df[df['actualDuration'] / df['responseTime'] <= 0.8]
    # afterWarmFilter = df.shape[0]
    # print(f"Warm start ratio: {(beforeWarmFilter - afterWarmFilter) / beforeWarmFilter}")

    print()

    print(path)
    print(df.sort_values(by='slowdown', ascending=False).head(10))

    print()

    return cdf(df, 'slowdown')


def plot_per_function_slowdown():