Multi-Leader Topologies: Star and Circular
What you'll learn: How star and circular topologies organize multi-leader replication to reduce network traffic at the cost of increased latency.
Beyond All-to-All: Alternative Topologies
In multi-leader replication, not every leader needs to talk directly to every other leader. While the all-to-all topology connects every leader to every other leader, star and circular topologies take different approaches to reduce network traffic and simplify connections—though they introduce trade-offs.
Star Topology
In a star topology, one central leader acts as the hub. All other leaders send their writes to this central node, which then forwards them to the remaining leaders.
Analogy: Think of a postal distribution center. Local post offices (outer leaders) send mail to the central hub, which sorts and distributes it to other local offices. No direct office-to-office shipping needed.
Trade-off: This reduces the number of connections from O(n²) to O(n), cutting network overhead. However, every write must pass through the central node, adding an extra hop—increasing latency. If the central node fails, replication breaks entirely (single point of failure).
Circular Topology
In a circular topology, leaders are arranged in a ring. Each leader forwards writes to the next leader in the circle, and writes propagate around until they reach all nodes.
Analogy: Like passing a note around a classroom circle—each person reads it and passes it to their neighbor until everyone's seen it.
Trade-off: This also reduces connections to O(n), but writes must traverse multiple hops to reach distant nodes, compounding latency. A single node failure can break the ring unless you implement redundant paths.
When to Use These
Use star or circular topologies when network bandwidth is constrained and you can tolerate higher propagation latency. They're rare in modern systems because all-to-all provides faster convergence with today's network speeds.
Key Takeaway: Star and circular topologies reduce replication traffic by routing through fewer connections, but they increase write propagation latency and introduce potential single points of failure.