Overview
The Koala benchmark suite enables characterization of performance-oriented research targeting the POSIX shell. It consists of 18 real-world program sets from diverse domains ranging from CI/CD and AI/ML to biology and the humanities. They are accompanied by real inputs that facilitate small- and large-scale performance characterization and varying opportunities for optimization. The latest version of Koala expands the suite with four new benchmark sets and extends its evaluation to new systems, testing zsh as an alternative shell interpreter and INCR, an incrementalization layer for shell programs.
Papers Using Koala
- SysXCHG: Refining Privilege with Adaptive System Call Filters — CCS 2023
- Rt: Regular Types for the Streaming Shell — OSDI '26
- Incr: Faster Re-execution via Bolt-on Incrementalization — OSDI '26
- hS: Speculative Script Reordering at Subprocess Granularity — OSDI '26
- Fractal: Fault-Tolerant Shell-Script Distribution — NSDI '26
- Controlling Opaque-Component Effects with Semisolates and Try — OSDI '26
If any aspect of the suite is useful, please use the following citation:
@inproceedings{koala2025atc,
title = {The Koala Benchmarks for the Shell: Characterization and Implications},
author = {Evangelos Lamprou and Ethan Williams and Georgios Kaoukis and Zhuoxuan Zhang
and Michael Greenberg and Konstantinos Kallas and Lukas Lazarek and Nikos Vasilakis},
booktitle = {Proceedings of the 2025 USENIX Annual Technical Conference (USENIX ATC '25)},
year = {2025},
address = {Santa Clara, CA},
publisher = {USENIX Association},
}Benchmarks
The complete list of benchmarks in the Koala suite is as follows:
| Name | Description |
|---|---|
analytics |
Real-world network logs filtering to extract and summarize key events. |
bio |
Genomic and transcriptomic analysis using population and RNA-seq data. |
ci-cd |
Open-source software project build and test. |
covid |
Analysis of public transit activity during the COVID-19 pandemic. |
etcetera |
Prime generation via a sieve and command execution inside an overlay sandbox. |
file-mod |
File compression, encryption, and conversion. |
inference |
Media-related inference tasks using large foundation models. |
interact |
Interactive shell programs (oh-my-zsh installer, snake game). |
ml |
A full machine learning pipeline using scikit-learn. |
net |
Network reconnaissance and firewall configuration tasks. |
nlp |
Book processing using shell-based NLP pipelines from Unix for Poets. |
oneliners |
Classic and modern one-liner shell pipelines. |
pkg |
AUR package build and npm packages permission inference. |
rand |
Random password generation and random team sampling from name lists. |
repl |
Security auditing and git development workflow replay. |
unixfun |
Text-processing problems from UNIX's 50-year anniversary. |
weather |
Analysis & visualization of historical weather statistics. |
web-search |
Crawling, indexing, and querying of Wikipedia data. |
Quick Setup
Koala can be obtained using the following ways:
- Run
curl up.kben.sh | shfrom your terminal, or - Clone the repo and run
cd koala && ./setup.sh, or - Fetch a Docker container by running
docker pull ghcr.io/kbensh/koala:latest, or - Build a Docker container from scratch.
More information in the README.
More Info
Recent News
Further Reading
- GitHub organization
- Short tutorial
- Contribution guide
Issues That Need Help
Community & More
- GitHub page: @kbensh
- Bug reports: koala/issues