Database Scaling: Sharding vs Replication

Question 1

Which technique is best for scaling a system with 100x more Reads than Writes?

Accepted Answer

Master-slave replication — Master-Slave Replication allows you to infinitely scale read throughput by adding more replicas, while keeping write logic simple.

Question 2

VERTICAL scaling (scaling up) means…

Accepted Answer

Upgrading one machine's hardware — Vertical scaling upgrades one machine; it's simple but bounded by hardware limits and a single point of failure.

Question 3

HORIZONTAL scaling (scaling out) means…

Accepted Answer

Adding servers to spread load — Horizontal scaling spreads work across many machines (replicas, shards), enabling near-limitless growth but adding complexity.

Question 4

READ REPLICAS scale…

Accepted Answer

Read throughput via copies — Replicas handle read traffic, offloading the primary; writes still go to the primary, so they don't scale write capacity.

Question 5

In primary-replica (master-slave) replication, writes go to…

Accepted Answer

The primary node only — A single primary accepts writes and propagates them to replicas that serve reads — read/write splitting.

Question 6

REPLICATION LAG is the delay…

Accepted Answer

Before a write hits replicas — Asynchronous replication means replicas trail the primary, so a read from a replica may return slightly stale data.

Question 7

To scale WRITES beyond one server, you typically need…

Accepted Answer

Horizontal sharding — Replicas don't increase write capacity; sharding splits writes across shards, each handling a portion of the data.

Question 8

CACHING (e.g. Redis) scales the database by…

Accepted Answer

Serving reads from memory — A cache absorbs repeated reads so the database handles fewer queries — often the cheapest first scaling step.

Question 9

Read/write SPLITTING routes…

Accepted Answer

Writes to primary, reads to replicas — Directing reads to replicas and writes to the primary distributes load — though apps must tolerate replica lag for reads.

Question 10

A limitation of VERTICAL scaling is…

Accepted Answer

A hardware ceiling limit — You can only add so much CPU/RAM to one box, and that box failing takes the database down — motivating horizontal approaches.

Master-Slave Replication

Replication or sharding?

Test your knowledge

Which technique is best for scaling a system with 100x more Reads than Writes?

Interview Questions

References & Further Reading

Related topics

Love this visualization?