All terms
Glossary
Auto scaling
Capacity follows load automatically, so that neither users wait nor idle resources get paid for.
Scaling is driven by signals such as utilization, queue length or requests per instance. Choosing the signal matters: CPU is rarely what users feel.
Scaling down belongs to it as much as scaling up, and that is the part that saves money and gets forgotten more often.
How you notice it
- Load varies noticeably across the day.
- Night time capacity matches peak capacity.
- The system was too slow during the last campaign.
Frequently asked
Why does my scaling react too late?
Because startup time exceeds the spike. Measure how long a new instance needs until it answers its first request. If that is longer than your spikes last, only pre warming or a faster start helps, no scaling rule will.