blksum can be up to 4 orders of magnitude faster than sha256sum,
depending on the contents of the image, image format, and the way we
read the image.
The most important factor is how much non-zero data is in the image. For images full of data, we are limited by storage throughput. For very sparse images we get dramatic improvement.
The following graphs compare the throughput using 4 images:
- 10p - 10% full image
- 20p - 20% full image
- 40p - 40% full image
- 80p - 80% full image
The benchmarks shown here ran on AWS c7g.metal instance with 64 cores running Ubuntu.
blksum performances is tested by running the blksum with real images
on local storage with various number of threads. This measures real
world performance of the command, which is typically limited by storage
throughput.
The real images for the benchmark are created using the
make-images.py tool. This tool requires the
virt-builder command and works only on Linux.
The results were generated by running bench/run tool:
bench/run bench/blksum.yaml
The graphs were created from the test results json files using the plot-blksum.py tool.
See blkhash performance to learn about the
maximum throughput possible using the blkhash library.
See the test/results directory for test results and images.