Scalability Tests Prediction from RadarGun Benchmark

scalability_test.py

import pandas as pd
from datetime import datetime
import numpy as np
# https://github.com/wip727/PyUSL/issues/1
from usl_v2 import usl_v2


def _append(node, max_index, row):
    sum_value = 0
    l = reversed(range(max_index + 1))
    for i in l:
        key = 'Slave ' + str(i)
        sum_value += row[key]
    master_key = 'Slave ' + str(max_index)
    if master_key not in node:
        node[master_key] = []
    node[master_key].append(sum_value)


nodes = range(8)

usecols = []
usecols.append('Timestamp')
for n in nodes:
    usecols.append('Slave ' + str(n))

data = pd.read_csv('/Users/dlovison/Downloads/timeline_asyncRepl_default_8_BasicOperations.Put_Throughput.csv',
                   usecols=usecols)
# convert Java timestamp to Python datetime
data['Timestamp'] = data['Timestamp'].apply(lambda x: datetime.fromtimestamp(x / 1000))
data.set_index('Timestamp', inplace=True)
data.fillna(0, inplace=True)

# lets sum every seconds
data_resample = data.resample('1s').sum()

node_dic = {}
for index, row in data_resample.iterrows():
    for n in nodes:
        if row['Slave ' + str(n)] != 0 and \
                ('Slave ' + str(n + 1) not in data_resample or row['Slave ' + str(n + 1)] == 0):
            _append(node_dic, n, row)
            break

y = []
x = []
for n in nodes:
    y.append(np.mean(node_dic['Slave ' + str(n)]))
    x.append(n + 1)

u = usl_v2()
# show current status
u.fit(x, y)
# predict 1 to 20 nodes
xgrid = np.linspace(1, 20, 200)
# show the prediction
u.plot(xgrid)

# I did a check with > peak.scalability(usl.dg) using R language and usl library

# source
# https://cran.r-project.org/web/packages/usl/vignettes/usl.pdf
# http://www.easysurf.cc/scintd.htm
# https://github.com/wip727/PyUSL
# http://www.perfdynamics.com/Manifesto/USLscalability.html#tth_sEc2.4
# https://www.vividcortex.com/resources/universal-scalability-law/
# https://www.vividcortex.com/resources/queueing-theory