AWS Cost-Conscious Architecture Study Guide

This guide covers the critical architectural strategies for cost optimization within AWS, focusing on purchasing models, resource rightsizing, and elastic scaling as defined in the SAP-C02 exam objectives.

Learning Objectives

After studying this guide, you should be able to:

• Differentiate between AWS purchasing models (On-Demand, Spot, RIs, and Savings Plans) based on workload characteristics.
• Apply rightsizing principles using AWS Compute Optimizer and S3 Storage Lens.
• Design elastic scaling strategies that align resource provision with real-time demand.
• Implement cost visibility and governance using tagging, AWS Budgets, and Cost Explorer.

Key Terms & Glossary

• Spot Instances: Spare compute capacity available at up to 90% discount, subject to reclamation by AWS with a 2-minute warning.
• Rightsizing: The process of matching instance sizes and types to your workload performance and capacity requirements at the lowest possible cost.
• Savings Plans: A flexible pricing model that offers low prices on EC2, Lambda, and Fargate usage in exchange for a commitment to a consistent amount of usage (measured in $/hour) for a 1 or 3-year term.
• Instance Fleets: A configuration for Auto Scaling or EMR that allows you to specify multiple instance types and purchasing options to increase availability and optimize cost.
• Cost Allocation Tags: Metadata assigned to resources (e.g., Project: Alpha) used to categorize and track AWS costs on a granular level.

The "Big Idea"

In traditional on-premises environments, "cost" is a capital expenditure (CapEx) handled during procurement. In AWS, Cost is a Performance Metric. An architect's goal is not just to build a functional system, but to build one that is "cost-optimized"—meaning every dollar spent contributes directly to business value. This requires moving from a "static provisioning" mindset to a "dynamic consumption" mindset.

Formula / Concept Box

Hierarchical Outline

1. AWS Purchasing Models
   • On-Demand: Highest flexibility, no commitment, highest cost. Best for new, unpredictable workloads.
   • Reserved Instances (RI): Commitment-based (1 or 3 years). Best for steady-state workloads.
   • Savings Plans: More flexible than RIs (applies across instance families). Best for evolving architectures.
   • Spot Instances: Up to 90% savings. Best for fault-tolerant, stateless, or batch workloads.
2. Rightsizing Strategies
   • Compute Optimizer: Uses ML to analyze historical metrics and suggest optimal EC2, EBS, and Lambda configurations.
   • Storage Tiering: Using S3 Intelligent-Tiering or lifecycle policies to move data to lower-cost tiers (Glacier) automatically.
3. Elasticity and Scaling
   • Horizontal Scaling: Adding/removing instances via Auto Scaling Groups (ASG) based on CloudWatch metrics.
   • Scheduled Scaling: Predicting known traffic spikes (e.g., Black Friday) to pre-provision capacity.
4. Cost Governance
   • Visibility Tools: Cost Explorer (trends), Trusted Advisor (idle resources), and AWS Budgets (threshold alerts).
   • Multi-Account Strategy: Using AWS Organizations and Consolidated Billing to maximize volume discounts.

Visual Anchors

Pricing Model Decision Tree

Demand vs. Capacity Optimization

Definition-Example Pairs

• Term: Stateless Workload
  • Definition: An application where no data is stored locally on the server; any instance can handle any request.
  • Example: A web fleet where session data is stored in ElastiCache/DynamoDB, allowing Spot Instances to be terminated without losing user data.
• Term: Storage Lens
  • Definition: A feature that provides organization-wide visibility into S3 storage usage and activity.
  • Example: Identifying buckets that have "non-current version" bloat and applying a lifecycle policy to delete them, saving thousands in monthly storage fees.

Worked Examples

Example 1: Converting Batch Jobs to Spot

Scenario: A company runs an ETL (Extract, Transform, Load) job every night that takes 4 hours on 10 m5.xlarge On-Demand instances ($0.192/hr each).

• Current Cost: $10 \times 4 \times 0.192 = $7.68$ per night.
• Spot Optimization: Converting to Spot Instances at a 70% discount.
• New Cost: $$10 \times 4 \times (0.192 \times 0.30) =$ $2.30$ per night.
• Implementation: The architect must ensure the ETL job can checkpoint its progress so that if an instance is reclaimed, the next one starts where the last left off.

Example 2: Rightsizing with Compute Optimizer

Scenario: An m5.2xlarge instance shows consistent CPU utilization of 5% and Memory usage of 10%.

• Analysis: AWS Compute Optimizer flags this as "Over-provisioned."
• Recommendation: Switch to a t3.medium or m5.large.
• Result: Monthly cost drops from ~$280 to ~$30, a 90% reduction for that specific resource.

Checkpoint Questions

1. Which pricing model is most appropriate for a baseline database server that must be available 24/7 for the next two years?
2. What is the minimum notification time AWS provides before reclaiming a Spot Instance?
3. How does AWS Lambda's pricing differ from EC2 in terms of granularity?
4. Which tool would you use to find idle Elastic Load Balancers (ELBs)?

Muddy Points & Cross-Refs

• RI vs. Savings Plans: RIs are often tied to specific instance types/regions (Standard RI), while Savings Plans apply automatically to any instance family, even if you switch from C5 to M5. Rule of thumb: Choose Savings Plans for compute flexibility.
• Spot Interruption: Many students fear Spot because of the 2-minute warning. Study Pointer: Look into "Spot Fleet" and "Capacity-optimized" allocation strategies to minimize the frequency of interruptions.

Comparison Tables

Comparison of Purchasing Models

[!IMPORTANT] Always prioritize Rightsizing before applying Savings Plans. If you commit to a 3-year plan for over-provisioned resources, you are "locking in" waste.