Technical

Compare architecture options before you commit

Give each architecture decision its own branch, compare the options side by side, and see which choices depend on each other.

Explore a reference map

The architecture of Redis on one map, zoomed out. Zoom in on any subsystem, such as persistence, clustering or the event loop, to read how it works.

Open the full map

How it works

Describe the system and the model maps its components. Each decision gets its own branch, and every question under it keeps the system’s requirements in context.

  1. Describe the requirements. Say what you are building. Ask for the main components, their boundaries and how they communicate.

  2. Branch on each decision. A question like “Should we use REST or GraphQL?” gets one branch per option, each with its tradeoffs.

  3. Follow the knock-on effects. Under an option, ask what it changes elsewhere. Choosing microservices, for instance, affects your deploy pipeline and how your teams are organized.

  4. Settle on a design. Delete or collapse the options that lose. What remains is a design you checked against the alternatives.

Why a map instead of a design doc

An Architecture Decision Record lists the alternatives to a choice, but it is written after the choice is made. A map lets you work through the alternatives before you decide, with each one’s consequences on its own branch.

Example

Say you are designing a notification system. The map branches into push or pull, real-time or batched, and per-user or topic-based delivery. Under real-time push you compare WebSockets, server-sent events and long polling. Under per-user delivery you look at fan-out strategies, and a follow-up there shows that real-time per-user delivery at scale needs a message broker.

Now try it yourself

Design the architecture for a real-time collaboration app
You'll need WebSocket-based communication, a CRDT or OT layer for conflict resolution, and a persistence layer that handles eventual consistency.
What are the trade-offs between CRDTs and OT?
CRDTs are simpler to reason about and work peer-to-peer, but use more bandwidth. OT is more established for rich text but requires a central server for transformation ordering.

Ready to try system architecture?