Notesfoundations🌱 seedling
Vertical Scaling
Also: Scaling Up · Scale Up · Shared-Memory Architecture
Vertical scaling adds capacity by moving to a more powerful machine: more CPUs, RAM and faster disks under one operating system. It keeps everything in one address space, so there are no distributed-systems problems, until you hit the biggest machine you can buy.
Why it exists
It's the cheapest scaling move in engineering effort: no code changes, no partitioning, no network between components, no partial failure.
How it works
- Shared-memory: one box, many cores, one OS. Today's ceiling in the cloud is around 1,920 vCPUs and 32 TiB of RAM (AWS U7inh, late 2024).
- Shared-disk is a halfway house: several machines with their own CPU and RAM share a storage array. It eases the ceiling but contends on locks.
- Usually paired with read replicas (Leader-Based Replication): one big primary takes writes, replicas absorb reads.
Trade-offs
| Gains | Costs |
|---|---|
| simple programming model, strong consistency for free | hard ceiling; you eventually run out of bigger boxes |
| no sharding, routing or cross-shard queries | one machine is one failure domain |
| upgrades usually mean a restart or failover |
In practice
- ChatGPT (2026): a single Azure PostgreSQL primary with ~50 read replicas served 800M users at millions of QPS.
- Figma (2020): one Postgres on AWS's largest physical instance, then vertical partitioning, then sharding when single tables neared the IOPS limit.
Pitfalls
- Waiting until you're at the ceiling to plan the next step; sharding takes months (Figma's first table took about nine).
- Confusing "fits on one machine" with "fault-tolerant": you still need a standby.
Connections
- Part of
- Scalability
- Contrasts with
- Horizontal Scaling
- Related
- Leader-Based Replication
- In real systems
- PostgreSQL
Recall
These cards are also in your review deck, scheduled with spaced repetition.