Question #1361
A media streaming company runs nightly video transcoding jobs on Amazon EC2 instances within an Auto Scaling group. The transcoding workload varies, but the baseline CPU utilization observed during each job is consistently 70%. The company must ensure capacity is provisioned 45 minutes before jobs begin. Currently, engineers manually adjust the Auto Scaling group's desired capacity, but the company lacks resources to analyze historical capacity trends. They require an automated solution with minimal operational overhead. Which approach best meets these requirements?
Implement a dynamic scaling policy targeting 70% CPU utilization. Configure the policy to adjust the Auto Scaling group's capacity based on real-time metric thresholds.
Set up a scheduled scaling policy with fixed desired, minimum, and maximum capacity values. Configure the recurrence to daily and set the start time 45 minutes before the transcoding jobs begin.
Deploy a predictive scaling policy using historical CPU utilization data. Configure the policy to forecast demand and launch instances 45 minutes prior to job execution, targeting 70% CPU utilization.
Use Amazon EventBridge to trigger an AWS Lambda function when CPU utilization reaches 70%. Configure the Lambda function to dynamically increase the Auto Scaling group's desired and maximum capacity by 25%.
Explanation
The correct answer is C. Predictive Scaling uses historical data to forecast demand and automatically provisions capacity in advance, aligning with the requirement to have instances ready 45 minutes before jobs begin. Since the company lacks resources to analyze trends, AWS's predictive scaling handles this automatically.
- Why not A/D: Dynamic scaling (A/D) reacts to real-time metrics (e.g., CPU utilization), which would scale instances after jobs start, failing the 45-minute pre-provisioning requirement.
- Why not B: Scheduled scaling (B) uses fixed capacity values, which may not adapt to varying workloads, leading to over/under-provisioning.
- Key Points: Predictive scaling combines historical trends and machine learning to forecast demand, ensuring timely scaling with minimal effort. It is ideal for recurring, time-based workloads with consistent patterns.
Answer
The correct answer is: C