Part 5 of 8 · 2 min read
A decision matrix
Compare the styles on the characteristics that matter.
A decision matrix lists the candidate styles in columns and the needed characteristics in rows. For each cell, note whether the style is strong, moderate or weak. Then look for the style whose weak cells fall on characteristics that matter least.
The table below gathers what pages 3 to 5 and the FSA style chapters say in words. It is a summary of tendencies, not star counts, and the book itself says ratings only point to tendencies.
| Layered / modular monolith | Pipeline | Microkernel | Service-based | Event-driven | Microservices | Space-based | |
|---|---|---|---|---|---|---|---|
| Deployment | One unit | One unit | One unit | Several coarse units | Several units | Many small units | Many processing units |
| Simplicity and cost | Strong | Strong | Strong | Moderate | Weak | Weak | Weak (costly, hard to test) |
| Scale and elasticity | Weak | Weak | Weak | Moderate | Strong | Strong | Strongest, built for it |
| Fault tolerance | Weak | Weak | Weak | Strong | Strong | Strong when calls between services are few | Strong |
| Deployability and agility | Weak (grows worse with size) | Weak | Moderate | Strong | Strong | Strong | Moderate |
| Testability | Moderate | Moderate | Moderate | Strong | Weak (many dynamic scenarios) | Strong per service | Very weak |
| Typical quanta | One | One | One | One or more | One to many | Many | Varies |
| Fits when | Small system, tight budget, unclear needs | One-way data processing | Product with plug-ins, per-client rules | Business apps wanting some distribution at low cost | Parallel reactions, high scale | Many teams, differing needs | Extreme variable load |
Where this page's table states a rating, it follows the words in the style chapters: layered strong on simplicity and cost, very low on deployability; microkernel strong on simplicity and cost, weak on scale and fault tolerance; service-based has no five-star rating but four stars in many areas; event-driven high on performance, scalability and fault tolerance but low on simplicity and testability; microservices high on scalability, elasticity and evolvability and low on performance; space-based expensive and the hardest to test. Pipeline is rated by its simplicity and one-unit deployment. Cells such as "Moderate" are a summary for this table and are not quoted from the book.
A worked example (made up). A company needs an internal expense app. Rank needs: simplicity and cost first, then maintainability. One set of characteristics fits all users. The matrix points to the layered or modular monolith. A video platform ranks scalability, elasticity and fault tolerance first, and has uneven load across parts. The matrix points to microservices or event-driven, with the cost in simplicity accepted.
Common mistake
Scoring every row as equal. A matrix helps only when the architect first picks the top three characteristics for this system (Level 0, page 5) and then looks at those rows.
Check yourself: 1. A team wants the cheapest option for a small app. Which row do you look at first? 2. Why is the "weak" cell for microservices on performance not a reason to rule them out? 3. Why must the top characteristics be chosen before reading the table?