Distributed systems are hard to test because partial failures overlap with live traffic and the same fault can produce different execution traces on different runs. Unmesh Joshi describes how deterministic simulation testing addresses this by replacing real clocks, networks, and storage with simulated components driven by a single-threaded execution loop. This allows simulating both hard failures and routine slowdowns—such as network and disk latency spikes—and more importantly makes every execution reproducible. This article explains the technique using a quorum key-value store as a running example, with code from Tickloom, a Java implementation of the approach.
Deterministic Simulation Testing
Find Your Interest
- So Darn Good
- Stencil Store Challenge
- Github and Open-source Is a Boon for the Underprivileged
- Github and Open-source Is a Boon for the Underprivileged
- Would you trust an AI bot to find the fix for vulnerabilities in your code?
- Just Visiting
- What's left for infrastructure-as-code after AI moves in?
- 实时预览敢让你敲 <script>:自研 Markdown 渲染器的「先转义后拼装」
- Applying Preact minification tricks to CSS
- Reflection





