CSE Roadmap

Core tier · about 2 hours · 6 parts

Part 3 of 6 · 1 min read

Then scale

Think of scale like choosing a vehicle for a move. A sedan, a van, and a moving truck can each carry boxes. Picking the wrong size either wastes money or cannot do the job at all.

In plain words: Once consistency needs are settled, the next question is how much data and throughput the workload actually has, and will have. Many workloads that feel like they need a specialized, horizontally scaled system actually fit comfortably on a single well-tuned machine running an ordinary relational or document database for years. A workload genuinely outgrowing a single machine's storage or throughput is the point where sharding, a distributed system, or a purpose-built high-throughput store like a wide-column database becomes justified, not before. Overestimating scale needs early adds real complexity, cross-shard joins, distributed transactions, that a single-machine system would never have required.

A worked example. The bookstore's catalog, even with a million books, fits comfortably on a single relational database instance for the foreseeable future. If the bookstore later ingests a continuous firehose of page-view events from millions of visitors, that specific workload, not the catalog, is the one that might actually justify a horizontally scaled store.

The key lines. No new SQL; this is a sizing judgment, not a query.

catalog:        ~1M rows, moderate read/write -> single instance is enough
page-view events: continuous high-volume stream -> may justify horizontal scale

Scale (workload)

Scale is the actual current and realistically projected data volume and request throughput a workload must handle, used to decide whether a single machine is sufficient or whether a horizontally distributed system is genuinely justified.

Example: the bookstore's catalog and its page-view event stream sit at very different points on this scale, even though both are "data the bookstore stores."