plot.py

#!/bin/env python3
# SPDX-License-Identifier: BSD-3-Clause
#
# Authors: Alexander Jung <alexander.jung@neclab.eu>
#s

import os
import csv
import sys
import fire
import numpy as np
from time import gmtime
from time import strftime
import matplotlib.pyplot as plt
import matplotlib.colors as mcolors
from common import sizeof_fmt, common_style, mk_groups, KBYTES, SMALL_SIZE, MEDIUM_SIZE, LARGE_SIZE
from os import listdir, makedirs

import pprint
pp = pprint.PrettyPrinter(indent=4)

def plot(data=None, output=None):
  WORKDIR = os.getcwd()
  RESULTSDIR = data
  RESULTEXT = '.csv'
  GROUP_BAR_WIDTH = .8
  DEFAULT = '_'
  THROUGHPUT = 'throughput'
  MEAN_KEY = 'mean'
  MEDIAN_KEY = 'median'
  AMAX_KEY = 'amax'
  AMIN_KEY = 'amin'

  files = []
  labels = []
  apps = []
  stats = {}
  throughput_max = 0 # maximum observed throughput
  total_apps = 0
  bar_colors = {
    'GET': '#FFF6F9',
    'SET': '#5697C4',
  }

  labels = {
    'mimalloc': 'Mimalloc',
    'tlsf': 'TLSF',
    'buddy': 'Binary buddy',
    'tinyalloc': 'tinyalloc'
  }

  for f in os.listdir(RESULTSDIR):
    if f.endswith(RESULTEXT):
      unikernel = f.replace(RESULTEXT,'')

      if unikernel not in stats:
        stats[unikernel] = {}
      
      with open(os.path.join(RESULTSDIR, f), 'r') as csvfile:
        csvdata = csv.reader(csvfile, delimiter="\t")
        
        next(csvdata) # skip header

        operations = {}

        for row in csvdata:
          if row[0] not in operations:
            operations[row[0]] = []
          
          operations[row[0]].append(float(row[1]) / 1000.0)
        
        for operation in operations:
          all_ops = np.array(operations[operation])
          operations[operation] = {
            MEAN_KEY: np.average(all_ops),
            MEDIAN_KEY: np.median(all_ops),
            AMAX_KEY: np.amax(all_ops),
            AMIN_KEY: np.amin(all_ops)
          }

          if int(round((np.amax(all_ops)))) > throughput_max:
            throughput_max = int(round((np.amax(all_ops))))

        stats[unikernel] = operations

  # General style
  common_style(plt)

  throughput_max += .4 # margin above biggest bar

  # Setup matplotlib axis
  fig = plt.figure(figsize=(8, 4))
  renderer = fig.canvas.get_renderer()

  # image size axis
  ax1 = fig.add_subplot(1,1,1)
  ax1.set_ylabel("Aver. Throughput (Million req/s)")
  ax1.grid(which='major', axis='y', linestyle=':', alpha=0.5, zorder=0)
  ax1_yticks = np.arange(0, throughput_max, step=0.5)
  ax1.set_yticks(ax1_yticks, minor=False)
  ax1.set_yticklabels(ax1_yticks)
  ax1.set_ylim(0, throughput_max)

  # Plot coordinates
  scale = 1 / (len(stats.keys()))
  xlabels = []

  # Adjust margining
  # fig.subplots_adjust(bottom=.15) #, top=1)

  i = 0
  line_offset = 0
  for unikernel in [
    'mimalloc',
    'tlsf',
    'buddy',
    'tinyalloc']:
    xlabels.append(labels[unikernel])
    operations = stats[unikernel]

    # Plot a line beteween unikernel applications
    if i > 0:
      line = plt.Line2D([i * scale, i * scale], [-.02, 1],
          transform=ax1.transAxes, color='black',
          linewidth=1)
      line.set_clip_on(False)
      ax1.add_line(line)

    j = 0
    bar_width = GROUP_BAR_WIDTH / len(operations.keys())
    bar_offset = (bar_width / 2) - (GROUP_BAR_WIDTH / 2)

    # Plot each application
    for operation_label in sorted(operations):
      print("mean " + str(operations[operation_label][MEAN_KEY]) + "max " + str(operations[operation_label][AMAX_KEY]) + "min " + str(operations[operation_label][AMIN_KEY]))
      bar = ax1.bar([i + 1 + bar_offset], operations[operation_label][MEAN_KEY],
        label=operation_label,
        align='center',
        zorder=4,
        yerr=(operations[operation_label][AMAX_KEY] - operations[operation_label][AMIN_KEY]),
        error_kw=dict(lw=2, capsize=10, capthick=1),
        width=bar_width,
        color=bar_colors[operation_label],
        linewidth=.5
      )
      
      ax1.text(i + 1 + bar_offset, operations[operation_label][AMAX_KEY] + 0.05, round(operations[operation_label][MEAN_KEY], 2),
        ha='center',
        va='bottom',
        zorder=6,
        fontsize=LARGE_SIZE,
        linespacing=0,
        bbox=dict(pad=-.6, facecolor='white', linewidth=0),
        rotation='vertical'
      )

      bar_offset += bar_width
      j += 1

    i += 1

  # sys.exit(1)

  # set up x-axis labels
  xticks = range(1, len(xlabels) + 1)
  ax1.set_xticks(xticks)
  # ax1.set_xticklabels(xlabels, fontsize=LARGE_SIZE, rotation=45, ha='right', rotation_mode='anchor')
  ax1.set_xticklabels(xlabels, fontsize=LARGE_SIZE)
  ax1.set_xlim(.5, len(xlabels) + .5)
  ax1.yaxis.grid(True, zorder=0, linestyle=':')
  ax1.tick_params(axis='both', which='both', length=0)

  # Create a unique legend
  handles, labels = plt.gca().get_legend_handles_labels()
  by_label = dict(zip(labels, handles))
  leg = plt.legend(by_label.values(), by_label.keys(),
    loc='upper right',
    ncol=2,
    fontsize=LARGE_SIZE,
  )
  leg.get_frame().set_linewidth(0.0)

  plt.setp(ax1.lines, linewidth=.5)

  # Save to file
  fig.tight_layout()
  #plt.show()
  fig.savefig(output) #, bbox_extra_artists=(ax1,), bbox_inches='tight')

if __name__ == '__main__':
  fire.Fire(plot)