Running a benchmark in Auto-pilot requires the following four steps:
- To run a benchmark, the benchmarker must write a configuration file that
describes which tests to run and how many times. The configuration file
does not describe the benchmark itself, but rather points at another
executable, which is usually a small wrapper shell script.
- The next step is to create the benchmark script itself. The script is
usually rather small, and provides arguments to a program like Postmark
or a compile benchmark. The wrapper script is also responsible for
We provide sample configuration files and shell scripts for benchmarking
file systems. These can easily be run directly for common file systems,
or easily adapted for other types of tests.
- Given the configuration file and the shell scripts, the next step is
to run the configuration file with Auto-pilot. Auto-pilot parses the
configuration file and runs the tests producing two types of logs.
The first type are simply the output from the programs. This can be
used to verify that benchmarks executed correctly and to investigate
any anomalies. The second log file is a more structured results file
that contains a snapshot of the system and the measurements that were
- The results file is then passed through our analysis program, Getstats,
to create a tabular report. Optionally, the tabular report can be used
to generate a bar or line graph using our plotting tools.
The rest of this manual is roughly divided into sections that correspond
to these actions. Configurations describes the Auto-pilot
configuration language. Scripts describes the scripts and hooks
that are included in the Auto-pilot distribution and Custom Scripts describes how to write your own scripts. Getstats
describes how to use and customize Getstats and Graphit describes
how to use our plotting tool.