Scaling writes: spread them, batch them, or shrink them
Reads can be copied and cached, but every write has to land somewhere durable, so scaling writes comes down to three moves (spread the writes across machines, batch them into fewer operations, or shrink them before they arrive), and each move is paid for in durability, latency or read complexity that you should name before the interviewer does.
What you'll be able to do
- Decide from a stated write rate whether one well-tuned node is enough before proposing anything distributed
- Quantify what batching buys in database operations and what it costs in latency and data at risk
- Choose between sharding, a queue, pre-aggregation and split counters by the shape of the write load
- State the durability and read-path cost of each technique in one sentence
Before this: back-of-the-envelope, picking-a-datastore, queues-and-backpressure
Log in to read the rest of this lesson
Reads can be copied and cached, but every write has to land somewhere durable, so scaling writes comes down to three moves (spread the writes across machines, batch them into fewer operations, or shrink them before they arrive), and each move is paid for in durability, latency or read complexity that you should name before the interviewer does. This walkthrough runs about 22 minutes, with runnable code you can edit and re-run as you read. A new account gets 7 days of Pro free, no card required.
Log in to continueThe first module of every track is free to read on the web: see what's open in System design.