EC2 Pricing Models — Complete Reference¶

1. Why Pricing Models Matter¶

AWS EC2 has multiple pricing models because workloads are not all the same. Choosing the wrong model = paying 3–10× more than necessary.

Core principle: Match the pricing model to workload behavior — not just to the cheapest price tag.

2. Billing Granularity ⭐ (Corrected)¶

Practical impact: A Linux/Windows instance running for 3 min 15 sec = billed for exactly 195 seconds — not 4 minutes, not 1 hour.

Windows is more expensive than Linux — not because of billing granularity, but because of the Microsoft license cost embedded in the per-second rate.

3. Complete Pricing Models Overview¶

4. On-Demand Instances¶

Pay for compute by the second. No commitment, no discount.

Characteristics: - No upfront payment - No minimum duration (beyond 60-second minimum billing) - Full flexibility — launch/stop/terminate anytime - Highest per-hour cost of all models

Use when: - Load is unknown or unpredictable - Short-term or temporary workloads - Dev/test environments - First-time running a new workload (before you know its pattern)

Engineering rule:

Flexibility needed > Cost savings → On-Demand

5. Reserved Instances (RI)¶

Commitment to a specific EC2 configuration for 1 or 3 years in exchange for discount.

5.1 RI Scope — Regional vs Zonal ⭐ (Often Missed)¶

Regional RI gives more flexibility. Zonal RI guarantees actual capacity — useful for critical workloads.

5.2 Offering Class¶

Standard RI = maximum savings, minimum flexibility. Convertible RI = lower savings, but you can exchange it if workload changes.

5.3 Payment Options¶

Pay more upfront → save more over the term.

5.4 RI Marketplace¶

Unused Standard RIs can be sold to other AWS accounts via the RI Marketplace — a way to recover cost if your workload changed unexpectedly.

Use when: - Workload is stable and predictable - Long-running baseline infrastructure (databases, app servers) - Cost optimization is a priority and usage is known 1–3 years ahead

6. Savings Plans (Modern, Flexible Alternative)¶

Commit to a minimum hourly spend ($/hr) for 1 or 3 years. AWS automatically applies discount to any eligible usage up to that committed amount.

6.1 Types of Savings Plans ⭐ (4 Types — Not 2 or 3)¶

AWS officially has four Savings Plan types as of 2025.

6.2 Compute vs EC2 Instance Savings Plan — Decision Guide¶

Are you running Lambda or Fargate alongside EC2?
    YES → Compute Savings Plan (covers all three)
    NO  → EC2 Instance Savings Plan (higher discount for pure EC2)

Are you locked into one instance family and Region?
    YES → EC2 Instance Savings Plan (up to 72%)
    NO  → Compute Savings Plan (up to 66%, full flexibility)

Key difference in numbers:

Workload: $10,000/month EC2 spend
Compute SP (66%):       → ~$3,400/month
EC2 Instance SP (72%):  → ~$2,800/month (if all in one family+Region)
Over 3 years:           → $21,600 difference

Use when: - You have a predictable minimum monthly compute spend - You want flexibility across instance families (Compute SP) - You're fully committed to one EC2 family and Region (EC2 SP)

7. Spot Instances ⭐¶

Unused AWS capacity sold at up to 90% discount. AWS can reclaim at any time.

How it works:

AWS has spare capacity in a pool (specific instance type + AZ)
You bid for that capacity
AWS gives it to you while available
If AWS needs it back → 2-minute interruption notice → instance terminated

Key rules: - Up to 90% cheaper than On-Demand - 2-minute warning before termination (via instance metadata + EventBridge) - Not available on-demand — depends on Region, AZ, and current demand - Cannot be converted to other pricing models

7.1 Spot Fleet¶

Request Spot capacity across multiple instance types and AZs in one request.

Spot Fleet Config:
  - c5.xlarge in us-east-1a
  - c5a.xlarge in us-east-1b
  - m5.xlarge in us-east-1a
  ...

Fleet uses allocation strategies to pick instances:

7.2 EC2 Fleet¶

Like Spot Fleet but supports mixed On-Demand + Spot + Reserved in a single request. Useful for auto-scaling groups that need guaranteed baseline + cheap burst.

Use Spot when: - Workload is fault-tolerant and stateless - Can checkpoint and resume (ML training, batch jobs) - Interruption = acceptable delay, not data loss

DO NOT use Spot for: - Databases - Session-critical applications - Any workload requiring guaranteed availability

8. Capacity Reservations¶

Reserve EC2 capacity in a specific AZ, billed at On-Demand rate — no discount.

Key property: Guarantees capacity exists when you need it.

Combining for best of both:

Capacity Reservation (guarantees the slot)
+
Regional RI or Savings Plan (applies discount to the billing)
= Guaranteed capacity + discounted price ✅

Use when: - Critical event with known capacity requirement (Black Friday, product launch) - Compliance requires guaranteed compute in specific AZ - Disaster recovery standby capacity

9. Dedicated Instances¶

Instances that run on hardware dedicated to your AWS account — no other customer shares the physical host.

• AWS controls which physical machine is used
• You have no visibility into cores, sockets, or host configuration
• Priced at a premium over standard On-Demand

Use when: Compliance or regulation requires single-tenant hardware.

10. Dedicated Hosts¶

A full physical server allocated exclusively to you.

Use when: - Bring Your Own License (BYOL) — Microsoft, Oracle, VMware require per-core licensing - Security audits require physical server visibility - Regulatory mandates full hardware isolation

11. Tenancy Summary¶

12. EC2 Status Checks ⭐¶

AWS runs automated health checks every minute:

Actions on failure: - System check failure → trigger instance recovery (moves to healthy host, same IP, EBS preserved) - Instance check failure → you must investigate (OS crash, OOM, disk full, misconfig)

# Check status via CLI
aws ec2 describe-instance-status --instance-ids i-xxxxxxxxxxxxxxxxx

13. Connection Methods¶

Session Manager (SSM) is the modern approach — no key pairs, no open ports, full audit trail via CloudTrail. Preferred for production.

14. Instance Identification¶

15. Cost Optimization Decision Framework¶

Step 1 — Is the workload interruptible?
    YES → Spot Instances (up to 90% off)
    NO  → continue to Step 2

Step 2 — Is usage predictable / long-term?
    NO  → On-Demand
    YES → continue to Step 3

Step 3 — Is it EC2-only or multi-service (Lambda/Fargate too)?
    EC2 only, single family → EC2 Instance Savings Plan (up to 72%)
    Multi-service or need flexibility → Compute Savings Plan (up to 66%)
    Need exact capacity guarantee → Capacity Reservation + Savings Plan

Step 4 — Does it need physical isolation?
    BYOL / license control → Dedicated Host
    Compliance only → Dedicated Instance
    Standard workload → Shared (default)

Scenario → Model Mapping¶

16. Common Mistakes ✅¶

17. Interview Questions Checklist ✅¶

• Name all EC2 pricing models
• What is the billing granularity for Linux? Windows? (per second — 60s min)
• On-Demand vs Reserved vs Savings Plans — when to use each?
• Standard RI vs Convertible RI — key differences?
• Regional RI vs Zonal RI — which one guarantees capacity?
• What are the 4 types of Savings Plans?
• Compute Savings Plan vs EC2 Instance Savings Plan — when to choose each?
• How does Spot Instance interruption work? How much warning?
• What is Spot Fleet? What allocation strategies are available?
• What is EC2 Fleet?
• What is a Capacity Reservation? Does it give a discount?
• How do you combine Capacity Reservation with Savings Plans?
• Dedicated Instance vs Dedicated Host — key differences?
• What is BYOL and which tenancy model supports it?
• What are the 3 EC2 status checks? Who fixes each failure type?
• How do you connect to a Linux EC2? Windows EC2?
• What is SSM Session Manager? Why is it better than SSH in production?
• Instance ID vs Instance Name — which can change?
• Walk through the cost optimization decision framework