Measure Power

This repository was created during the course Computer Systems Performance at the IT University of Copenhagen - by Andreas Trøstrup, Frederik Petersen, and Lucas Hanson.

It was created to compare the energy consumption of different programming languages, specifically C, Java, JavaScript, TypeScript, Ruby, and Zig.

The paper concluded that even though there is a tendency for compiled languages to use less energy it is hard to draw strict conclusions from the data. This is due to that implmentation, the transpiler chosen, the libraries used, and the hardware used all have a significant impact on the energy consumption of the program.

This program was run on a Raspberry Pi 5 with 8GB of RAM.

Language dependencies

Language	Compiler/Runtime	Build Flags	Runtime Features
C	gcc 12.2.0	-pipe -Wall -O3 -fomit-frame-pointer -march=native -fopenmp	Native execution
Java	javac 24.0.1 (GraalVM)	native-image --silent --gc=G1 -cp . -O3 -march=native	Native execution with GraalVM
JavaScript	N/A	N/A	Deno 2.3.3 (V8 13.7.152.6)
TypeScript	N/A	N/A	Deno 2.3.3 (TypeScript 5.8.3)
Ruby	N/A	N/A	Ruby 3.1.2p20 with YJIT
Zig	zig 0.14	-O ReleaseFast -lc	Native execution

Benchmark-specific dependencies

Benchmark	C	Java	JavaScript/TypeScript	Ruby	Zig
binarytrees	Standard C library	java.util.concurrent	Deno standard library	Ruby standard library	Zig standard library
fannkuchredux	OpenMP	java.util.concurrent	Deno standard library	Ruby standard library	Zig standard library
fasta	Standard C library	Java standard library	TextEncoder, Deno I/O APIs	Ruby standard library	Zig standard library
knucleotide	Custom hash table (khash.h), OpenMP	Java collections, concurrency	TextDecoder, Deno I/O APIs	Ruby standard library, parallel execution	Zig standard library, custom Map implementation
mandelbrot	OpenMP	java.util.concurrent	Deno I/O APIs, navigator.hardwareConcurrency	Ruby standard library	Zig standard library
nbody	Standard C library, math library	Java standard library	Deno standard library	Ruby standard library	Zig standard library
pidigits	GMP library (-lgmp)	java.math.BigInteger	JavaScript BigInt	Ruby standard library	Zig standard library
regexredux	PCRE2 library (-lpcre2-8)	java.util.regex, java.util.concurrent	JavaScript RegExp, Deno I/O APIs	Ruby Regexp	Zig standard library
reversecomplement	Standard C library	Java standard library	TextDecoder, Deno I/O APIs	Ruby standard library	Zig standard library
spectralnorm	Standard C library, math library	Java standard library	Deno standard library	Ruby standard library	Zig standard library

Running the benchmarks

There are multiple steps needed to run the benchmarks. The measure_power.py script is used to measure the power consumption of the program. But it was created to run while having a Prometheus server running in the background - so that the power consumption can be measured over time. To get Prometheus to work, please follow the instructions on their website here.

Please look at Language Dependencies and Benchmark-specific dependencies for more information on the dependencies used to run each benchmark.

You also need to create an input folder with txt files generated by running fasta with:

• 100000001
• 5000000
• 25000000

and name them, respectively:

• revcomp-input100000001.txt
• regexredux-input5000000.txt
• knucleotide-input25000000.txt

To run the measure_power.py script, run the following commands (we have only run it with Python 3.11):

python3 -m .venv venv
source venv/bin/activate
pip install -r requirements.txt

Then run the script with the following command:

python3 measure_power.py

A csv file will be created named energy_results.csv.

Visualizing the results

Run the following command to create visualizations, and a LaTeX table of the results:

python3 visualize.py

This will create the following files:

• build_energy_comparison.png
• relative_run_energy.png
• run_energy_comparison.png
• total_energy_comparison.png
• energy_usage_tables.tex