MLS-C01 Question 278
Select 3You are deploying a machine learning model on Amazon EC2 instances using an Auto Scaling group. The model requires GPU instances for inference. To minimize costs while ensuring availability, you want to use a combination of On-Demand and Spot Instances. How should you configure the Auto Scaling group to meet these requirements?
- A
Define a launch template with a mixed instance policy that includes both On-Demand and Spot Instances.
- B
Set the desired capacity in the Auto Scaling group to 100% On-Demand Instances to ensure consistent performance.
- C
Use EC2 Fleet or Spot Fleet to manage the Spot Instances separately from the Auto Scaling group.
- D
Configure instance weighting in the Auto Scaling group to prioritize GPU instances based on their availability and cost.
- E
Enable capacity-optimized allocation strategy to ensure Spot Instances are provisioned from the most available pools.
Show answer and explanation
Correct answers: A, D, E
Explanation
To deploy a cost-effective and highly available Auto Scaling group for a GPU-based ML model, you should use a mixed instance policy to combine On-Demand and Spot Instances. Additionally, configuring instance weighting ensures the group can balance availability and cost across instance types. Enabling the capacity-optimized allocation strategy further reduces the risk of Spot Instance interruptions by selecting the most available Spot capacity pools.
- A. Correct.
Defining a launch template with a mixed instance policy allows you to specify both On-Demand and Spot Instances within the Auto Scaling group, enabling cost optimization while maintaining availability.
- B. Incorrect.
Setting the desired capacity to 100% On-Demand Instances ensures consistent performance but does not optimize costs, which is a key requirement in this scenario.
- C. Incorrect.
Using EC2 Fleet or Spot Fleet separately is unnecessary since Auto Scaling groups support mixed instance policies and can manage both On-Demand and Spot Instances effectively.
- D. Correct.
Configuring instance weighting allows you to assign different weights to instance types, ensuring the Auto Scaling group can meet capacity requirements while optimizing for cost and availability.
- E. Correct.
The capacity-optimized allocation strategy ensures that Spot Instances are provisioned from pools with the most available capacity, reducing the likelihood of instance interruption.