Clbg Energy Eval

clbg-energy-eval

The Green Side of the Lua — André Brandão et al. (arXiv:2601.16670, 2026)

What this evaluates

Evaluates the runtime performance and energy efficiency of different programming language implementations. It specifically compares Lua interpreters, LuaJIT JIT compilers, and C on computationally intensive benchmark programs from the Computer Language Benchmarks Game.

Datasets

• CLBG (Computer Language Benchmarks Game) — total ?; splits: test (-1)

Metrics

• Energy Consumption (primary) — range: other
  • Total energy measured in Joules (J) using Intel RAPL for CPU Package and DRAM domains.
• Runtime — range: other
  • Execution time measured in seconds.
• Speedup — range: other
  • Ratio of runtime of non-optimized version ($T_{\phi}$) to optimized version ($T_{o}$). Values > 1 indicate faster optimized implementation.
• Greenup — range: other
  • Ratio of energy consumption of non-optimized version ($E_{\phi}$) to optimized version ($E_{o}$). Values > 1 indicate reduced energy usage.
• Powerup — range: other
  • Ratio of power consumption of optimized version ($P_{o}$) to non-optimized version ($P_{\phi}$), calculated as Speedup / Greenup. Values < 1 indicate lower average power consumption.

Input / output format

Input: CLBG benchmark programs (e.g., fannkuch-redux, spectral-norm, Fibonacci) implemented in Lua, LuaJIT, or C, executed on a laptop with Intel RAPL energy monitoring.

Output: Raw energy consumption (Joules for Package and DRAM domains), execution time (seconds), and derived ratios (Speedup, Greenup, Powerup) relative to a reference implementation.

Scoring recipe

def compute_green_metrics(runtime_phi, runtime_o, energy_phi, energy_o):
    speedup = runtime_phi / runtime_o
    greenup = energy_phi / energy_o
    powerup = speedup / greenup
    return {'Speedup': speedup, 'Greenup': greenup, 'Powerup': powerup}

Common pitfalls

• Energy measurements rely on Intel RAPL, which may not capture all system power draw or vary across hardware.
• Comparisons across different programming language rankings use different machines/environments, making cross-study comparisons unreliable.
• Outliers are removed using the Inter-Quartile Range (IQR) method before analysis, which can skew results if not reported transparently.

Evidence (verbatim from paper)

Considering two different versions of a specific software, one that is optimized and one that is not, the following metrics are defined:

In the Speedup equation $T_{\phi}$ represents the runtime of the non-optimized version, and $T_{o}$ represents the runtime of the optimized version. This metric quantifies the performance gain, where values above 1 indicate a faster optimized implementation.

Citation

@misc{brandao2026greensideoflua,
  title={The Green Side of the Lua},
  author={André Brandão et al.},
  year={2026},
  note={arXiv:2601.16670}
}

• arXiv: 2601.16670