dd block size

The dd POSIX utility is useful for copying bytes between files. The bytes are read and written in blocks whose size is controlled by the bs parameter.

The default value of bs is 512 bytes, presumably to match the historical customary block size of hard drives. Is it a good default on modern computers?

• 
• 

import subprocess
import tempfile
import re
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt

reps = 10
pattern = r'(?P<time>\d\.\d*) s'
bss = (2**np.arange(10, 27, 2)).astype(int)
timings = []

# set up dummy files to copy bytes to and from
tmp1 = tempfile.NamedTemporaryFile('w')
tmp2 = tempfile.NamedTemporaryFile('r')
io_files = [
    ('/dev/zero', '/dev/null'),
    (tmp1.name, tmp2.name)
]
subprocess.run(['dd', 'if=/dev/zero', f'of={tmp1.name}', 'bs=1M', 'count=64'])

for input_file, output_file in io_files:
    for bs in bss:
        count = int(bss[-1] / bs)
        cmd = ['dd', f'if={input_file}', f'of={output_file}', f'bs={bs}', f'count={count}']
        for i in range(reps):
            text = subprocess.run(cmd, capture_output=True).stderr.decode()
            m = re.search(pattern, text)
            timings.append({
                'bs': bs,
                'time': float(m.groupdict()['time'])
            })

    data = pd.DataFrame(timings)
    data.plot.scatter('bs', 'time', logx=True, logy=True)
    plt.ylabel('Copy Time [s]')
    plt.xlabel('Block Size [byte]')
    plt.title(rf'{input_file} $\rightarrow$ {output_file}')

tmp1.close()
tmp2.close()

previous

Pandas efficiency

next

Correlation efficiency