Cloud auto-scaling dynamically adjusts compute resources based on demand, allowing applications to maintain performance during traffic spikes while minimizing costs during low-utilization periods. This capability has evolved from simple threshold-based reactions into sophisticated predictive systems using machine learning that anticipate load changes before they occur. Understanding the distinction between horizontal scaling (adding instances) and vertical scaling (increasing instance size), along with when to apply reactive versus proactive strategies, determines whether your infrastructure scales efficiently or burns budget fighting fires.
What This Cheat Sheet Covers
This topic spans 14 focused tables and 116 indexed concepts, 108 flashcards. Below is a complete table-by-table outline of this topic, spanning foundational concepts through advanced details.
A jump-to index of every table row in this cheat sheet.
An interactive map of every table and concept in this topic.
Table 1: Core Scaling Approaches
The choice between horizontal and vertical scaling, and between reactive and proactive strategies, is the most fundamental decision in any auto-scaling design — get it wrong and you either over-provision constantly or chase demand with lag.
| Strategy | Example | Description | |
|---|---|---|---|
Add 3 web servers min=2, max=10 | • Increases capacity by adding more instances to distribute workload • provides fault tolerance and theoretically unlimited scalability but requires stateless design or external session management | ||
t3.medium → t3.xlarge 2 vCPU → 4 vCPU | • Increases capacity by upgrading to a larger instance type • simpler implementation with no architectural changes but hits hardware limits and typically requires instance replacement | ||
CPU > 70% for 5 min → add instance | • Responds to observed metrics after load increases • simple to configure and prevents over-provisioning, but introduces lag between demand surge and new capacity being ready | ||
Scale at 08:00 daily Based on forecast | • Adds capacity before anticipated load using schedules or ML predictions • eliminates reactive lag but risks over-provisioning if forecasts are inaccurate | ||
ML forecasts next 48h Pre-provisions capacity | Uses machine learning on 14+ days of historical traffic to forecast future load and provision resources ahead of anticipated surges — now available in all major AWS regions. | ||
Min=10 at 09:00 Mon-Fri Min=2 at 18:00 | • Adjusts capacity at predetermined times based on known traffic patterns • ideal for business-hours workloads or batch windows where demand cycles are predictable | ||
Auto-detect demand Scale 0 to 1000s | • Platform automatically provisions and de-provisions resources with no manual intervention • serverless platforms achieve this natively while container orchestrators use controllers to match desired state | ||
Start with 4×large Scale out to 8×large | • Combines both approaches by first scaling up to larger instances then scaling out when vertical limits are reached • optimizes resource density while preserving horizontal expansion capability |