simple-disk-benchmark

A simple disk benchmark tool.

Operating Systems

Currently, macOS and Linux are tested. Windows support is limited, unit tests pass, but the software is not otherwise tested or validated on Windows. Development is on macOS.

Installation

From homebrew (macOS only):

brew tap schwa/schwa
brew install simple-disk-benchmark

Or with cargo (all platforms):

cargo install simple-disk-benchmark

Usage

A simple disk benchmark tool

Usage: simple-disk-benchmark [OPTIONS] [FILE]

Arguments:
  [FILE]  File to use for benchmarking. If this file exists it will be deleted [default: testfile.dat]

Options:
  -s, --size <FILESIZE>         Size of the file to use for benchmarking [default: 1GB]
  -b, --blocksize <BLOCK_SIZE>  Size of the blocks to read/write [default: 128MB]
  -c, --cycles <CYCLES>         Number of test cycles to run [default: 10]
  -m, --mode <MODE>             Types of test to run: read, write or all [default: all] [possible values: all, read, write]
  -r, --random-seek             Seek to a random position in the file before each read/write
      --no-create               Do not create the test file, the file must already exist
      --no-delete               Do not delete the test file after the test
      --no-progress             Do not display progress bar
      --no-disable-cache        Do not disable the file system cache
      --no-close-file           Do not close the file after each cycle
      --no-random-buffer        Fill the buffer with fixed byte pattern on creation instead of random
  -X, --no-chart                Do not display a bar chart of the run timings
  -j, --export-json <FILE>      Export the timing summary statistics and timings of individual runs as JSON to the given FILE. The output time unit is always seconds
      --export-log <FILE>       Export the log to the given FILE
  -d, --dry-run                 Do not actually perform benchmarks to the disk (file is still created and/or deleted)
  -v, --verbose...              Increase logging verbosity
  -q, --quiet...                Decrease logging verbosity
  -h, --help                    Print help
  -V, --version                 Print version

Methodology

Specify the path to a file to use for benchmarking. If the file exists, it will be deleted. You can specify paths on other drives by using the full path to the file (e.g. /Volumes/MyDrive/testfile.dat).

The benchmark tool will create a file of the specified size and then run the specified number of cycles. Each cycle will read or write the multiple blocks of a specified size to the file.

On macOS, the file is opened and F_NO_CACHE and F_GLOBAL_NOCACHE are both set on the file descriptor. This will bypass the file system cache and write directly to the disk. On Linux, the O_DIRECT flag is used to achieve the same result.

TODO

• Display volume info in the preamble.
• Better output - display timing info as well as rates.
• Multithreaded benchmarking option.
• More documentation.
• Run test coverage.
• Borrow CPU time stuff from hyperfine.
• Fuzz the StyleSheet code.
• Put on homebrew.
• Separate file creation from opening for runs.
• Test on Windows.
• Random seeks instead of just sequential.
• Use random bytes instead of zeros for writes
• Use a better ByteSize replacement.
• Add a --no-delete option to keep the file around after the benchmark.
• Output data to JSON.
• Output data to CSV. Won't do this. Use JSON and pipe to jq or csvkit.
• Find out what's going on with all the dead_code false positives.
• Sort out pub/mod stuff.

License

MIT License. See LICENSE file.