Elasticity in cloud computing is the ability of a system to add and remove resources automatically as demand changes, both up and down. NIST calls this rapid elasticity. It lets you match capacity to current load, so you neither pay for idle servers nor fall over during a spike.
How elasticity works
Elastic systems follow a simple loop:
- Monitor: The platform tracks metrics such as CPU use, memory, network traffic or request counts.
- Trigger: When a metric crosses a threshold you set, a scaling rule fires.
- Scale: Resources are added or removed. New instances or containers start, or existing ones are resized.
- Balance: A load balancer spreads traffic across the healthy resources.
Elasticity is automatic, but it has to be set up. You define the metrics, thresholds and limits. Without them, nothing scales.
Elasticity vs scalability
People often use these words interchangeably, but they describe different things.
| Elasticity | Scalability | |
|---|---|---|
| What it is | Automatic adjustment up and down with demand | The ability to handle growth over time |
| Time frame | Short-term, moment to moment | Long-term capacity planning |
| How it happens | Usually automatic, through rules and metrics | Often planned, sometimes with manual steps |
| Example | Extra servers during a flash sale, then removed | Upgrading infrastructure as the customer base grows |
In short, scalability is about capacity to grow. Elasticity is about matching capacity to demand as it rises and falls. Elasticity is one of the most useful forms of scalability.
Types of elasticity
Vendor and industry material describes three approaches:
- Reactive elasticity: Resources change in response to demand that has already happened. This avoids paying for resources you do not need, but there can be a short delay before new capacity arrives.
- Proactive elasticity: Resources are added in advance, based on forecasts or a schedule. This can reduce the delay, but it can cost more if the forecast is wrong.
- Manual elasticity: A person watches demand and adjusts resources. This is the slowest approach and removes much of the benefit.
Where you will see it
- AWS: EC2 Auto Scaling groups add and remove virtual machines according to policies you configure.
- Microsoft Azure: Virtual Machine Scale Sets, with Azure Monitor autoscale rules.
- Google Cloud: Managed instance groups with an autoscaler.
- Kubernetes: The Horizontal Pod Autoscaler checks metrics every 15 seconds by default and adjusts the number of pods. A cluster autoscaler or a node provisioner such as Karpenter adds or removes the nodes that run them.
- Serverless: Services such as AWS Lambda scale with incoming requests and can scale to zero when idle.
A worked example: an online store on a busy weekend
An online store usually needs modest capacity. On a holiday sales weekend, traffic may be several times higher for a few days.
- Without elasticity: The store buys enough servers for the peak and pays for them all year, or it under-provisions and the site slows or fails during the sale.
- With elasticity: Scaling rules add instances as CPU or request counts rise, and remove them as traffic falls. The store pays for the extra capacity only while it is used.
Limits and trade-offs
Elasticity is powerful but not unlimited:
- Startup time: A virtual machine can take minutes to start, so scaling does not react instantly to a sudden spike.
- Quotas: Providers set limits. For example, AWS Lambda has a default concurrency quota of 1,000 executions per account per region, which can be raised.
- Configuration: Poorly chosen thresholds can scale too late, scale too early or scale without limit, which can raise the bill.
- Application design: Applications that store session data or files on one server do not scale cleanly. They need redesign, such as moving state to shared storage.
- Security: New instances must get the same access controls and network protection as the rest of the system.
Set a maximum capacity and a billing alert before you enable automatic scaling. Without a cap, a bug or an attack can scale up costs quickly.
Common misconceptions
- Elasticity is unlimited. Quotas and regional capacity still apply.
- Scaling is instant. New virtual machines need time to boot and warm up.
- Elasticity and scalability are the same. They are related, but they describe different time frames.
- Elasticity happens automatically. You have to define the metrics, rules and limits.
Frequently asked questions
What is elasticity in cloud computing in one sentence?
It is the ability of a cloud system to automatically add and remove resources to match demand, in both directions.
Is elasticity the same as auto scaling?
Auto scaling is the mechanism most providers use to deliver elasticity. Elasticity is the outcome: capacity that follows demand.
Does elasticity save money?
It can. You pay for capacity while you use it instead of for peak capacity all year. The savings depend on how much your demand varies, and on keeping scaling limits and alerts in place.
For more on how scaling works in practice, read scalability in cloud computing. For the other essential characteristics, see key characteristics of cloud computing.