Examkin

Question

Difficulty: Very hardCost-Optimized Compute Selection and Purchasing Strategies

A media processing company is designing a system on AWS to orchestrate and run video rendering jobs. The architecture includes:

1. An orchestration daemon that runs 24/724/7 on Amazon Elastic Container Service (Amazon ECS) on AWS Fargate to monitor job queues and manage metadata.
2. A fleet of worker nodes that retrieve tasks from an Amazon SQS queue, download large video segments, process them, and upload the results to Amazon S3. Individual rendering tasks can run for up to 3 hours3\text{ hours}. The queue depth varies significantly throughout the day, and tasks can be safely restarted from the beginning if interrupted.
3. A persistent database layer running on Amazon RDS for PostgreSQL to store customer metadata and job configurations.

Which combination of compute selection and purchasing strategies will achieve the lowest overall cost? (Select TWO.)

  1. Purchase a Compute Savings Plan to cover the persistent orchestration daemon running on AWS Fargate.Answer
  2. Provision the worker node fleet using Amazon EC2 Spot Instances to process the batch video rendering tasks.Answer
  3. C
    Purchase a Compute Savings Plan to cover both the orchestration daemon on AWS Fargate and the Amazon RDS for PostgreSQL database instance.
  4. D
    Deploy the video rendering tasks on AWS Lambda functions to run the rendering jobs serverlessly.
  5. E
    Configure the worker nodes in a private subnet and route all video downloads and uploads to Amazon S3 through a NAT Gateway.

Answer

The most cost-effective compute strategy is to purchase a Compute Savings Plan for the persistent 24/724/7 Fargate orchestration daemon, and to use Amazon EC2 Spot Instances for the fault-tolerant, batch video rendering worker nodes.
The optimal cost-effective strategy uses a Compute Savings Plan for the persistent Fargate orchestration daemon, providing substantial discounts for 24/724/7 workloads. For the batch rendering worker fleet, Amazon EC2 Spot Instances are selected because the tasks are stateless, can be restarted from the SQS queue upon interruption, and run for up to 3 hours3\text{ hours}, allowing the company to save up to 90%90\% compared to On-Demand pricing.

Step-by-Step Solution

1
Analyze the execution profile of the orchestration daemon.
The orchestration daemon runs continuously (24/724/7) on AWS Fargate.
Since the resource usage is predictable and persistent, a commitment-based model like a Savings Plan is appropriate to reduce costs compared to On-Demand rates.
2
Determine the eligibility of services for Compute Savings Plans.
Compute Savings Plans apply to AWS Fargate but do not apply to Amazon RDS instances.
A database-specific purchasing model or Reserved Instance would be needed for the RDS PostgreSQL database, whereas the Fargate compute can be covered by the Compute Savings Plan.
3
Analyze the execution profile of the video rendering worker fleet.
The worker fleet processes stateless, fault-tolerant batch workloads from a queue. Individual tasks can run for up to 3 hours3\text{ hours} and can be safely restarted if interrupted.
Because the workload is interruptible and spiky, EC2 Spot Instances provide the deepest discount (up to 90%90\% off On-Demand) and fit the fault-tolerant nature of SQS-based processing.
4
Evaluate the suitability of AWS Lambda for the rendering tasks.
AWS Lambda is unsuitable because its maximum execution timeout is 15 minutes15\text{ minutes}, whereas the rendering tasks run for up to 3 hours3\text{ hours}. Lambda is also not cost-optimized for long-running, continuous compute tasks.
Confirming that Lambda is mathematically and technically restricted from handling this workload.
5
Evaluate the networking costs of the worker nodes accessing Amazon S3.
Routing S3 data transfers through a NAT Gateway incurs high data processing fees. Using a free S3 Gateway VPC Endpoint bypasses the NAT Gateway.
Minimizing data transfer costs is a critical part of compute architecture design.

Key Concept

Matching workloads to cost-optimized AWS compute hosting models and purchasing strategies (Spot Instances for fault-tolerant batch processing, and Compute Savings Plans for persistent, predictable containerized workloads on AWS Fargate).
Rate this question

Topics

Cost-Optimized Compute Selection and Purchasing Strategies