What we time
Timing covers the whole command, including startup and connection setup. Creating the test files and checking their contents happen outside that time. These tests finish when the tool exits; they do not include a final flush to persistent storage.
What the charts show
Copy speed is the file size divided by the average time across runs. Longer bars mean faster copies. The unchanged-folder test shows elapsed time instead: shorter bars are better, and no file contents need transferring.
The thin line spans the fastest and slowest runs. Hover, tap or focus the bar to see each measurement. Speeds use decimal units: 1 MB is 1,000,000 bytes.
For updates to existing files, speeds use the size of the whole folder. Tools may reuse old contents, so this is a way to compare how quickly they finish the job, rather than a measure of network traffic.
Making a fair comparison
Each tool gets the same files and starting destination. We repeat the job three times per tool, vary which tool goes first, and verify every copy with checksums. Release charts require every tool to complete all three runs successfully, with each run lasting at least three seconds.
The setup depends on the test. Local-copy tests use disks; server-to-server tests keep files in memory to focus on the network. Filesystems, caching and hardware can affect the outcome.
Each comparison on All results includes its setup, individual runs and downloadable measurements. For instructions, see Reproduce a result.