AWS Compute Purchasing Options: The SAA-C03 Study Guide

This guide covers the various ways to purchase and pay for Amazon EC2 and other compute resources. Selecting the right purchasing option is critical for the "Cost Optimization" pillar of the AWS Well-Architected Framework.

Learning Objectives

After studying this guide, you should be able to:

• Compare and Contrast On-Demand, Reserved Instances (RIs), Savings Plans, and Spot Instances.
• Select the most cost-effective model based on workload predictability and flexibility requirements.
• Identify the differences between Standard and Convertible RIs.
• Explain the mechanics of Spot Instance interruptions and bid pricing.

Key Terms & Glossary

• Tenancy: The physical relationship between your instance and the underlying hardware (Shared, Dedicated Instance, or Dedicated Host).
• Commitment: A contractual agreement to pay for a specific amount of usage over a 1-year or 3-year term.
• Underutilization: When a reserved resource is not being used, but the customer is still being charged the commitment rate.
• Interruption: The process where AWS reclaims a Spot Instance with a 2-minute warning because the capacity is needed elsewhere.

The "Big Idea"

The core philosophy of AWS purchasing is a trade-off between flexibility and cost. On-Demand instances offer maximum flexibility but the highest price. By committing to a term (Reserved/Savings Plans) or accepting volatility (Spot), you can achieve significant discounts.

[!TIP] Always think of On-Demand as the "Baseline" and determine how much of that baseline can be moved into commitment-based or spot-based models to save money.

Formula / Concept Box

Hierarchical Outline

• 1. On-Demand Instances
  • Low Commitment: Pay by the second; no long-term contracts.
  • Fixed Price: Highest cost but guaranteed availability.
• 2. Reserved Instances (RI)
  • Payment Options: All Upfront (cheapest), Partial Upfront, No Upfront.
  • Types:
    • Standard: Max discount; can only change AZ within a region.
    • Convertible: Lower discount; allows changing instance family/OS.
    • Scheduled: Reserved for specific recurring time windows.
• 3. Savings Plans
  • Compute Savings Plan: Applies to EC2, Fargate, and Lambda regardless of region or family.
  • EC2 Instance Savings Plan: Region-specific; requires commitment to an instance family.
• 4. Spot Instances
  • Capacity-Based: AWS sells unused capacity at a steep discount.
  • Spot Price: Fluctuates based on demand; if it exceeds your max bid, the instance terminates.

Visual Anchors

Decision Flow: Choosing a Purchasing Option

Cost Comparison Visualization

Definition-Example Pairs

1. Compute Savings Plan

   • Definition: A flexible pricing model that provides low prices on EC2, Fargate, and Lambda usage in exchange for a commitment to a consistent amount of usage (measured in $/hour).
   • Example: A company runs a web app on EC2 in US-East-1 but plans to migrate the backend to AWS Lambda in 6 months. A Compute Savings Plan will apply the discount to both during the transition.

2. Spot Fleet

   • Definition: A collection of Spot Instances (and optionally On-Demand instances) that allows you to maintain a target capacity by automatically requesting the cheapest available instances across different pools.
   • Example: A genomic sequencing lab needs 1000 CPUs for a 48-hour run. They use a Spot Fleet to grab various instance types (m5.large, c5.xlarge) at 90% off, knowing that if one pool is reclaimed, the fleet will request replacements from another pool.

Worked Examples

Problem 1: The 24/7 Database

Scenario: A company has a production database running on a t3.large instance. It must be available 24/7 for at least the next two years. The instance family is unlikely to change.

• Solution: Standard Reserved Instance (3-year term).
• Reasoning: Since the workload is 24/7 (steady-state) and predictable, the 3-year Standard RI offers the maximum discount (up to 75%).

Problem 2: Batch Image Processing

Scenario: A social media app processes user-uploaded photos to create thumbnails. The processing can happen anytime within 4 hours of the upload and is handled by a worker queue.

• Solution: Spot Instances.
• Reasoning: The workload is stateless and fault-tolerant (if an instance is terminated, the job just stays in the queue for the next available instance). Spot provides the lowest cost.

Checkpoint Questions

1. What is the main difference between a Compute Savings Plan and an EC2 Instance Savings Plan?
2. If a Spot Instance is terminated by AWS, do you pay for the partial hour of usage? (Note: AWS policy varies by termination cause).
3. Which payment option for Reserved Instances results in the highest total cost?
4. Can a Standard RI be converted to a different instance family mid-term?

▶Click to see answers

1. Compute Savings Plans apply to EC2, Fargate, and Lambda across any region/family. EC2 Instance Savings Plans are restricted to a specific instance family in a specific region.
2. If AWS terminates a Spot Instance for capacity, you are not charged for the partial hour. If you terminate it yourself, you are charged for the full hour.
3. No Upfront is the most expensive; All Upfront is the cheapest.
4. No. Only Convertible RIs allow changing instance families. Standard RIs only allow changing Availability Zones within the same region.