test_cpu_parallel – CPU & Memory Load Generator in Rust

test_cpu_parallel is a small, zero‑dependency load‑testing kit. It can:

• load cores to verify scaling or system/container limits
• sweep 1 → N threads and export a tidy CSV for speed‑up plots
• hammer the memory hierarchy with random‑access touches beyond the LLC
• run stand‑alone from the CLI or be embedded in your own Rust code

A two‑minute build, no external data files, works on Linux, macOS and Windows.

Link to the Project page

📦 Installation

Note Note The tool intentionally disables compiler optimizations using an unoptimized build for cargo build --release and cargo install (by setting opt-level = 0). While cargo build also generates an unoptimized binary for debug, it is not equivalent to a --release build. For consistent performance comparisons, avoid mixing cargo build binaries with --release binaries.

⚡ CLI quick‑start

# Burn four threads for a few seconds
$ test_cpu_parallel -w 4

# Sweep 1 → 16 threads, write CSV
$ test_cpu_parallel --num_workers 16 --full -o results.csv

# Stress memory with 8 workers and a 512 MiB buffer
$ test_cpu_parallel -w 8 --mode memory --memory_size 512

Run test_cpu_parallel --help for the complete flag list (excerpt below).

Usage: test_cpu_parallel [OPTIONS]
  -w, --num_workers <N>        parallel threads      (default 2)
  -m, --mode <cpu|memory>      workload type         (default cpu)
  -f, --full                   sweep 1‥=N threads
  -o, --output_file <PATH>     write CSV (full mode)
      --memory_size <MiB>      buffer for memory test (default 1024)

📚 Library usage

Embed the engine in unit tests, benchmarks or monitoring agents:

use test_cpu_parallel::TestCPUParallel;

fn main() -> anyhow::Result<()> {
    // 4 workers, one batch, memory workload, 64 MiB buffer
    let bench = TestCPUParallel::new(4, 1, 1_000, "", 64);
    let (thread_stats, batch_stats) = bench.test_one_load(None, "memory")?;
    println!("Per‑thread  stats: {thread_stats:?}");
    println!("Per‑batch   stats: {batch_stats:?}");
    Ok(())
}

API docs are hosted on docs.rs/test_cpu_parallel.

🔧 Building from source

# Clone and build a debug binary (recommended for comparable timings)
$ git clone https://github.com/LucaCanali/Miscellaneous.git
$ cd Miscellaneous/Performance_Testing/Test_CPU_parallel_Rust
$ cargo build --release

📊 Analysing the CSV

Full‑mode output is a single CSV line per thread count, ready to feed into Python/pandas, gnuplot or Excel. See the Notebooks/ directory in the repo for ready‑made Jupyter notebooks that plot speed‑up and efficiency curves.

More

• Project page & extended docs – https://github.com/LucaCanali/Miscellaneous/tree/master/Performance_Testing/Test_CPU_parallel_Rust
• Blog article that motivated the tool – https://db-blog.web.cern.ch/node/189
• Licence – Apache‑2.0

Enjoy hacking your CPUs 🔥