Arkinu
AWS service comparison

AWS Auto Scaling vs Amazon EC2 Auto Scaling: differences and when to use each

AWS Auto Scaling
Amazon EC2 Auto Scaling

Often confused on the examQuestions tend to offer one as a distractor for the other: look for the phrase in the scenario that tips the balance.

Management and Governance

AWS Auto Scaling

Service that monitors applications and automatically adjusts their capacity to maintain steady performance. With scaling plans, it configures scaling for multiple resources from a single place, such as Amazon EC2 Auto Scaling groups, Amazon ECS tasks, Amazon DynamoDB tables, or Amazon Aurora replicas.

Exam phrases that point here

  • «configure scaling for multiple application resources (EC2, ECS, DynamoDB, Aurora) from a single place»

    AWS Auto Scaling creates a scaling plan that covers Auto Scaling groups, ECS services, Spot Fleets, DynamoDB tables, and Aurora replicas at the same time, with consistent policies across the whole stack.

    When not: If only EC2 instances are being scaled, if you have to create the Auto Scaling group, or if you need step or scheduled scaling policies, the answer is Amazon EC2 Auto Scaling; AWS Auto Scaling plans use only target tracking.

  • «choose a scaling strategy that prioritizes availability, cost, or a balance of the two»

    Scaling plans offer recommended strategies (optimize for availability, optimize for cost, or balance) that set the target utilization value for each resource, along with suggested minimum and maximum limits.

    When not: If each resource needs its own metrics, thresholds, or very specific scaling steps, you configure scaling directly in each service (EC2 Auto Scaling or Application Auto Scaling).

  • «have capacity ready before recurring daily or weekly traffic peaks»

    Predictive scaling uses machine learning on the load history, forecasts the next few days, and schedules capacity increases before the peak arrives.

    When not: In scaling plans, predictive scaling applies only to EC2 instances. If you only want predictive scaling, AWS recommends configuring it as a policy directly on the Auto Scaling group; and a one-time event with a known date is handled with a scheduled action.

  • «discover an application's scalable resources from its CloudFormation stack or its tags»

    When you create a plan, AWS Auto Scaling scans the CloudFormation stack or the resources with certain tags and lists the ones that can be scaled.

    When not: ECS services are not discovered by tags, and the stack search does not traverse nested stacks. If the scenario asks you to create the infrastructure, that is CloudFormation, not AWS Auto Scaling.

  • «automatically adjust the read and write capacity of a DynamoDB table based on usage»

    A scaling plan can increase and decrease the provisioned capacity of a table or a global secondary index, together with the application's other resources.

    When not: If only that table needs to be scaled, configuring its scaling in DynamoDB (Application Auto Scaling) is enough. AWS Auto Scaling adds value when several resources of the same application are coordinated.

Often paired with

  • Amazon ECS
  • Amazon DynamoDB
  • Amazon Aurora
Study AWS Auto Scaling →

Compute

Amazon EC2 Auto Scaling

Maintains the right number of Amazon EC2 instances for your application's load by using Auto Scaling groups with a minimum, maximum, and desired capacity. It launches or terminates instances according to scaling policies and automatically replaces unhealthy instances.

Exam phrases that point here

  • «automatically adjust the number of EC2 instances based on demand»

    An Auto Scaling group adds instances when load rises and removes them when it drops, so you pay only for the capacity you need.

    When not: If what needs to scale is not EC2 instances but, for example, Amazon ECS tasks or the capacity of a DynamoDB table, the answer lies in the scaling features of those services. And if the scenario asks not to manage servers, think of AWS Lambda or AWS Fargate.

  • «keep the group's average CPU utilization close to 50%»

    A target tracking policy sets a metric and a target value; Auto Scaling creates the CloudWatch alarms and adjusts the number of instances to get close to that value, like a thermostat.

    When not: If work arrives through an Amazon SQS queue, the number of messages in the queue does not work directly for target tracking: you use a messages-per-instance metric. And if you want fine-grained control over how many instances to add at each threshold, a step scaling policy fits.

  • «automatically replace instances that stop responding»

    Auto Scaling checks the health of instances (the EC2 status checks and, if enabled, the load balancer's) and replaces failing ones to maintain the desired capacity.

    When not: The replacement creates a new instance from the template: anything stored only on the old instance is lost. If the data must survive, it belongs outside the instance (Amazon RDS, Amazon DynamoDB, Amazon S3, Amazon EFS).

  • «high availability of the application tier across multiple Availability Zones»

    An Auto Scaling group can span multiple Availability Zones in a Region and distributes instances among them as evenly as possible.

    When not: An Auto Scaling group does not span multiple Regions. If the scenario asks to withstand the failure of an entire Region, you need resources in another Region and a mechanism such as Amazon Route 53 to direct traffic. And for the database, think of Amazon RDS Multi-AZ, not Auto Scaling.

  • «a base capacity that cannot be interrupted and the rest of the capacity at the lowest cost»

    A single group can combine purchasing options: a base capacity covered by On-Demand Instances (with an RI discount if it matches) and anything above that by Spot Instances or a mix.

    When not: If no part of the workload tolerates interruptions, everything must be On-Demand (with Savings Plans or RIs to lower the cost). Spot fits only the part that can lose instances without harming the service.

Often paired with

  • Elastic Load Balancing (ELB)
Study Amazon EC2 Auto Scaling →