This repository contains the implementation of my Bachelor's thesis on optimizing Function as a Service (FaaS) performance.
It introduces a technique that benchmarks cloud instances during the function’s initialization phase, utilizing idle time (e.g., during data downloads) to assess performance. If an instance is too slow, the function intentionally fails and retries on a new instance. This ensures that execution only continues on fast-performing instances, leading to an average speedup of 5.24% and up to 8.71% on the best day.
Below is an overview of the project structure and the most important directories to help you navigate the codebase.
⚠️ Note: This project has been taken down and is no longer executable.
It is published for reference purposes only — primarily to provide access to the collected data and logs.
Contains the Google Cloud Functions used in the project:
-
function-baseline.go
The baseline function used for performance comparison. -
function-optimized.go
The optimized function that implements the performance improvement strategy. -
function-execute-benchmark.go
A helper function that measures how long benchmarks typically take.
Jupyter notebooks and logs for experimentation and analysis:
-
logs_analysis/
Contains daily log files collected during the experiment. -
workflow-1/
Notebooks for executing the experiment workflow. -
workflow-2/
Notebooks for analyzing the experimental results.
Contains scripts for running load tests using k6.
We executed the experiment twice to ensure a backup in case of issues. Since the first run completed without problems, only the data from the first execution was used in our analysis.