MLS-C01 Question 288
Select 2You are deploying a machine learning model on AWS that needs to handle real-time inference requests with low latency. During peak traffic, you notice uneven distribution of requests to the endpoints hosting your model, resulting in some instances being overloaded while others are underutilized. Which of the following solutions would help evenly distribute the traffic across all instances?
- A
Configure an Application Load Balancer (ALB) in front of your endpoints and enable sticky sessions.
- B
Use Amazon SageMaker Endpoint Autoscaling to dynamically adjust the number of instances based on traffic patterns.
- C
Implement weighted routing in Amazon Route 53 to distribute traffic evenly across multiple endpoints.
- D
Leverage an Application Load Balancer (ALB) with a round-robin load balancing strategy to distribute requests across your endpoints.
- E
Use AWS Global Accelerator to direct traffic to the nearest endpoint with the least latency.
Show answer and explanation
Correct answers: B, D
Explanation
To resolve uneven traffic distribution, combining Amazon SageMaker Endpoint Autoscaling with an ALB using a round-robin strategy is an effective solution. Autoscaling dynamically adjusts the number of instances based on traffic, and the ALB ensures requests are distributed evenly across all active instances. Sticky sessions and weighted routing are not suitable for solving this load balancing issue, and AWS Global Accelerator focuses on latency-based routing rather than even distribution.
- A. Incorrect.
Enabling sticky sessions on an ALB will cause requests to stick to a particular instance, which could lead to uneven load distribution. This does not solve the problem mentioned in the scenario.
- B. Correct.
Amazon SageMaker Endpoint Autoscaling is a recommended solution to handle variable traffic patterns by dynamically adjusting the number of active instances, which helps prevent overloading.
- C. Incorrect.
Weighted routing in Amazon Route 53 could help distribute traffic based on weights, but this does not ensure even distribution across endpoints during real-time traffic surges.
- D. Correct.
Using an ALB with a round-robin load balancing strategy will evenly distribute incoming requests across all instances, resolving the issue of uneven traffic distribution.
- E. Incorrect.
AWS Global Accelerator is used for directing traffic to endpoints based on latency and availability, but it does not ensure even traffic distribution across all instances.