Included Targets

SemaTune is organized around application targets. A target is any workload or service that can be prepared, measured over a time window, and summarized into metrics. The repository includes two maintained targets that are used throughout the docs and tests.

The config field is still named benchmark for compatibility. On this page, benchmark: "sysbench_cpu" and benchmark: "tpcc" mean “select this target adapter.”

sysbench_cpu

sysbench_cpu runs the built-in sysbench CPU prime-number workload. It is the fastest target to use when checking installation, validating OS parameter application, or learning the control loop.

Typical objective:

{
  "optimization_metric": "throughput",
  "optimization_goal": "maximize"
}

Important target fields are documented in the Configuration Reference, and the step-by-step run guide is Running Sysbench CPU.

tpcc

tpcc runs BenchBase TPC-C against PostgreSQL. It is the database-backed example target and exercises a more realistic setup path: PostgreSQL service configuration, BenchBase build/load, per-window execution, and tail-latency metrics.

Typical objective:

{
  "optimization_metric": "latency_p99",
  "optimization_goal": "minimize"
}

Important target fields are documented in the Configuration Reference, and the setup guide is Running TPCC.

Adding More Targets

These included targets are examples of the target-adapter interface, not the limit of the framework. To add another application, implement the adapter contract described in Extending SemaTune: prepare the target, observe or run one measured window, parse metrics, clean up, and provide a small example config plus tests.