In the SAA-C03 exam, cost-optimized compute makes up roughly 18% of the questions. The question is almost always the same: "Which purchase option gives the lowest cost for this type of workload?" Understanding the differences through everyday analogies makes this section straightforward.
What Are EC2 Purchase Options?
Think about airline tickets. The same seat on the same plane can cost wildly different prices: a deeply discounted advance-purchase ticket (Reserved Instances), a full-price same-day ticket (On-Demand), a standby last-minute bargain (Spot Instances), or an annual frequent-flyer membership (Savings Plans). AWS EC2 works the same way — the same computing resource is available at different prices depending on how you commit to it. Choose the right option and you can cut costs by up to 90%.
On-Demand Instances
Why does it exist? Sometimes you genuinely cannot predict when or how much you will use. On-Demand lets you use resources without any upfront commitment.
What is it? No prepayment, no contract, start when you need it, pay only for the hours (or seconds) you run. Think of a convenience store: full price, but available 24/7 with no strings attached.
How does it work? The moment you start an instance, billing begins per second. The moment you stop it, billing stops.
When to use it: New services where demand is unpredictable Short-term testing or development environments lasting a week or two Workloads that cannot be interrupted but where long-term commitment is uncertain
Reserved Instances
Why does it exist? If you know you will be running a workload continuously for a year or more, you can get up to 72% off by making a commitment in advance.
What is it? A 1-year or 3-year contract on a specific instance type and region. Think of a 1-year mobile phone contract: your monthly bill is lower, but breaking the contract early comes with a penalty.
How does it work? Three payment options exist: All Upfront (pay everything now, maximum discount), Partial Upfront (pay some now, lower monthly rate), or No Upfront (pay monthly, smallest discount). The more you pay upfront, the lower your total cost.
Example: A production web server running 24/7/365 is an ideal candidate for Reserved Instances. Up to 72% savings compared to On-Demand.
Spot Instances
Why does it exist? AWS data centers always have unused server capacity. Rather than let it sit idle, AWS sells that spare capacity at a steep discount. Everyone wins: AWS monetizes idle resources, and customers get massive savings.
What is it? Like a standby airline ticket — extremely cheap (up to 90% off), but AWS can take those resources back with only 2 minutes of warning if another customer needs them. The key word is interruptible.
How does it work? You set a maximum bid price. When the current Spot price is below your bid, your instance runs. If the Spot price rises above your bid, you get a 2-minute interruption notice and then your instance is stopped or terminated.
When to use it: Batch data processing jobs that can be retried Rendering, machine learning training, genomics analysis — fault-tolerant workloads Spot Fleet: automatically manages a pool of Spot Instances across multiple instance types and Availability Zones
Warning: Never use Spot Instances for production databases or any workload that cannot be interrupted.
Savings Plans
Why does it exist? Reserved Instances are powerful, but they lock you into a specific instance type and region. Savings Plans give you the same discount level with much more flexibility.
What is it? Think of a gym membership. You pay a fixed monthly fee and can use any equipment in any location within the chain. With Savings Plans, you commit to a minimum hourly spend ($/hr) and get discounted pricing on any eligible compute usage within that spend.
How does it work? You commit to spending at least a certain dollar amount per hour (for example, $10/hr). Usage up to that committed amount is billed at the discounted Savings Plans rate. Any usage beyond that is billed at On-Demand prices.
| Item | Savings Plans | Reserved Instances | |------|-------------|-------------------| | Flexibility | High (change instance type and region) | Low (type and region are locked) | | Discount | Up to 72% | Up to 72% | | Scope | EC2, Fargate, Lambda | EC2 only | | Commitment | Hourly spend amount ($/hr) | Number of instances |
Tip: If you expect to change instance types or regions, choose Savings Plans over Reserved Instances.
Spot + On-Demand Mixed Strategy (Auto Scaling)
Why does it exist? You want stable baseline capacity with cheap burst capacity for peak traffic.
How does it work? In an Auto Scaling group, configure your baseline instances as On-Demand (or Reserved), and set additional scaling instances as Spot. Traffic spikes are absorbed cheaply by Spot Instances. If a Spot Instance gets reclaimed, the On-Demand baseline keeps your service running.
Right-Sizing and AWS Compute Optimizer
Why does it exist? When engineers first provision instances, they tend to choose large ones "just to be safe." In practice, many instances run at 20–30% CPU utilization, and the rest is wasted money.
What is it? AWS Compute Optimizer analyzes CloudWatch metrics from your running instances and recommends the right size. It is like a mechanic telling you: "Your car has a 3.0L engine but you only use city roads — a 1.6L engine would save you fuel with no performance loss."
What it optimizes: EC2 instance types and sizes EBS volume sizes Lambda memory allocations ECS on Fargate task sizes
Switching to Serverless to Change Your Cost Structure
Why does it exist? EC2 instances charge you even when they are idle. Serverless charges you only when code is actually running, so idle time costs almost nothing.
Lambda: billed per request count and execution duration in milliseconds. Scales automatically to thousands of requests per second.
Fargate: billed only for container runtime. No EC2 instances to manage.
Example: A report-generation task that runs for 5 minutes per day costs 24 hours of EC2 billing if left on all day. With Lambda, you pay only for those 5 minutes.
Exam Key Points
"Interruptible workload, up to 90% savings" -- Spot Instances
"Like a standby ticket — cheap but can be reclaimed" -- Core characteristic of Spot Instances
"On-Demand baseline + Spot additional capacity" -- Auto Scaling mixed strategy
"Flexible discount, applies to EC2, Fargate, and Lambda" -- Savings Plans
"Fixed discount, locked to instance type and region, 1 to 3 year term" -- Reserved Instances
"Plan to change instance types or regions frequently" -- Choose Savings Plans over Reserved Instances
"Downsize an over-provisioned instance to the right size" -- AWS Compute Optimizer
"Eliminate idle-time charges, pay only when running" -- Serverless (Lambda and Fargate)
"Bring your own existing software license to a physical server" -- Dedicated Hosts
Never use Spot Instances for production databases or non-interruptible workloads