repobench(1) Manual Page

In generate mode (the default) it runs reposurgeon multiple times with different readlimit values and writes out performance statistics for each run to a datafile. The name of that datafile is reported.

In display (-p) mode, a benchmark-data file is sent to gnuplot to make a graph. An alternate display mode (with -o) ships the graph to standard output as an SVG.

Generate mode takes four arguments, the dump file to read followed by the minimum, step, and maximum values to use for readlimit. The dump file may be specified as a path relative to the current directory. This leaves the following files in the current directory:

The data will be recorded in a file named after the current git revision, as an aid to comparing multiple historical revisions of the code.

New data is simply appended to the end of the data file, so it is useful to run this script multiple times with different step values. For example, the GCC repository had ~280k revisions. Running this first with a step of 50k to see a rough graph in a shorter amount of time, followed by runs with smaller step values, such as 10k or 1k, is recommended.

Generate mode uses sudo to instruct the kernel to drop everything from the file system cache between runs.

To change the behavior of reposurgeon you can user the -c option to pass it a command-line argument.

Display mode with a single file argument takes a previously generated datafile and renders it as a graph. Curve fitting is attempted. Use the -p option for interactive gnuplot, or the -o option to ship an SVG to stdout. The -s option can be used to choose which statistic to graph.

Giving multiple files with display mode suppresses curve fitting and produces a comparative graph. The -o, -p, and -s options are interpreted. This mode is useful in checking for performance regressions.