Examkin

Soru

Zorluk: Çok zorCost-Optimized Compute Selection and Purchasing Strategies

A healthtech startup has migrated its core application platform to AWS and wants to establish the most cost-optimized compute purchasing strategy. The application workload consists of:

* An API tier running on Amazon ECS using AWS Fargate that requires a continuous, predictable baseline of 1.50perhourincomputecosts,24/7.∗AdatabasetierrunningonAmazonRDSforMySQLthatrequiresacontinuous,predictablebaselineof1.50 per hour in compute costs, 24/7. * A database tier running on Amazon RDS for MySQL that requires a continuous, predictable baseline of 2.00 per hour in database compute costs, 24/7.
* A daily data transformation process running on Amazon ECS using AWS Fargate that is highly fault-tolerant, runs for 2 hours every night, and can scale up to 50 tasks dynamically.
* An ad-hoc analytics training model running on Amazon EC2 GPU instances for exactly 6 hours every Saturday, which must run to completion without interruption.

Which strategy meets these requirements at the lowest cost?

  1. Purchase a Compute Savings Plan to cover the baseline Fargate API tier. Purchase Amazon RDS Reserved Instances to cover the database tier. Use Fargate Spot for the daily data transformation tasks, and use On-Demand Instances for the weekly EC2 GPU analytics training.Cevap
  2. B
    Purchase a Compute Savings Plan to cover both the baseline Fargate API tier and the Amazon RDS database tier. Use Fargate Spot for the daily data transformation tasks, and use On-Demand Instances for the weekly EC2 GPU analytics training.
  3. C
    Purchase an EC2 Instance Savings Plan to cover both the baseline Fargate API tier and the weekly EC2 GPU analytics training. Purchase Amazon RDS Reserved Instances to cover the database tier, and use Fargate Spot for the daily data transformation tasks.
  4. D
    Purchase a Compute Savings Plan to cover the baseline Fargate API tier. Purchase Amazon RDS Reserved Instances to cover the database tier. Use AWS Lambda to run the daily data transformation tasks to leverage serverless billing, and use On-Demand Instances for the weekly EC2 GPU analytics training.

Cevap

Purchase a Compute Savings Plan to cover the baseline Fargate API tier, purchase Amazon RDS Reserved Instances for the database tier, use Fargate Spot for the daily data transformation tasks, and use On-Demand Instances for the weekly EC2 GPU analytics training.
The correct strategy leverages the appropriate discount mechanism for each service type and workload characteristic. A Compute Savings Plan is selected for the baseline AWS Fargate API tier. Amazon RDS Reserved Instances are selected for the database tier since database compute is excluded from Compute Savings Plans. Fargate Spot is chosen for the daily data transformation process as it is fault-tolerant and containerized. On-Demand instances are chosen for the weekly GPU analytics training because its low duty cycle (6 hours per week) makes a 24/7 Savings Plan commitment economically inefficient, and its requirement to run to completion without interruption rules out Spot instances.

Adım Adım Çözüm

1
Analyze the API tier requirements.
The API runs on AWS Fargate 24/7. Since this is a continuous, predictable compute workload, a Compute Savings Plan is the ideal purchasing model, offering up to 66% savings over On-Demand.
Baseline 24/7 compute workloads benefit most from Savings Plans.
2
Analyze the database tier requirements.
The database runs on Amazon RDS 24/7. Compute Savings Plans do not cover Amazon RDS. Instead, RDS Reserved Instances must be purchased to secure a discount on this steady-state database workload.
Savings Plans and Reserved Instances have distinct scopes.
3
Analyze the daily data transformation process.
The process runs on Fargate for 2 hours daily and is highly fault-tolerant. Fargate Spot is the most cost-effective option here, providing up to 70% discount compared to standard Fargate.
Fault-tolerant, containerized batch workloads are a perfect fit for Spot capacity.
4
Analyze the weekly GPU analytics training.
The training runs on EC2 GPU instances for only 6 hours a week and cannot be interrupted. Because it cannot be interrupted, Spot Instances cannot be used. Because it runs for only 6 hours out of 168 hours in a week, committing to a Savings Plan would result in paying for 162 hours of idle time. Therefore, On-Demand is the most cost-effective choice.
Intermittent, non-interruptible workloads with low weekly utilization should use On-Demand rather than commitments.

Anahtar Kavram

Selecting cost-optimized compute types and purchasing models based on workload predictability, execution duration, and service type.
Tahmini Süre:3m 0s
Bu soruyu puanla

Konular

Cost-Optimized Compute Selection and Purchasing Strategies