Machine Learning helps AWS determine when spikes in resource demands will occur
Predictive scaling is a new feature designed for AWS Auto Scaling groups to make enterprise cloud environments more efficient. With AWS Auto Scaling, AWS can track and adjust your compute resources to scale with a sudden rise in demand. This gives customers a scalable and flexible approach when using EC2.
With the new predictive scaling, customers no longer have to manually configure Auto Scaling. Instead, the service will use algorithms to analyze daily and weekly patterns. It then predicts demand and automates the provisioning of EC2 instances ahead of the spikes. Note, however, that this works for regularly occurring surges, not ones that suddenly and randomly appear within that time frame.
Overall, predictive scaling makes AWS Auto Scaling groups more efficient, responsive, and cost-effective. It may augment or, in some cases, replace a user’s scaling policies.
Using Predictive Scaling
Navigate to AWS Auto Scaling console. Click “Get Started”. From here, select which resources you want to scale. You may choose AWS Auto Scaling groups, CloudFormation stacks, or use tags to search for items.
From here, you will want to formulate a scaling plan. You may optimize it for cost, availability, or a mixture of the two. Once the predictive scaling box is checked, the feature is enabled. If you want, you may also activate dynamic scaling, which employs target-tracking policies determined by factors you have set in your Auto Scaling group. Note that predictive scaling needs at minimum one day of data to create predictions. Each day, it will re-evaluate the data and develop a prediction for the next 2 days.
Cloud admins may choose not to activate predictive scaling in a production environment. Regardless, the analytics for predicting surges can be used for testing and analyzing your workload’s resource consumption.
Predictive scaling is most useful for clients who need to anticipate scaling instead of just reacting to changes. It is also useful for reducing cloud operation costs by controlling how it scales instances. This lets enterprises automate scaling of compute resources for no added cost.
This service is currently available in the following regions: US. East (North Virginia, U.S. East (Ohio), U.S. West (Oregon), Europe (Ireland) and Asia Pacific (Singapore). If your company uses AWS Auto Scaling groups, consider predictive scaling now.