Pricing models: On-Demand, Reserved Instances, Savings Plans, Spot and Dedicated options, data transfer and storage tiers

Why this matters

The same EC2 instance, running the same software, can cost very different amounts depending on how it was bought. A company that runs everything at the default rate, keeps archives in its most expensive storage class and sends data between Regions without thinking about it is paying for convenience it does not need. Choosing the right purchasing option is often the largest single saving available, and it changes nothing about how the application works.

A manager approving a cloud budget needs to know which commitments are on offer and what each one gives up in return. A salesperson needs to explain why a three-year plan is cheaper than paying as you go, and when it is the wrong choice. A junior engineer is often the first person to notice that a batch job could run on Spot, or that a bucket of old logs belongs in an archive tier.

This lesson teaches the models and the trade-offs, not the prices. Prices differ by Region and change over time, so look up current rates on each service’s pricing page and model them in the AWS Pricing Calculator, covered in lesson 24.

The lesson

Compute purchasing options: On-Demand, Reserved, Savings Plans, Spot, Dedicated and Capacity Reservations

Every option below runs the same Amazon EC2 instances. What differs is the commitment you make, the discount you get, and whether AWS can take the capacity back.

• On-Demand Instances. You pay for compute only while the instance runs, with no commitment and no upfront payment. AWS does not reclaim them for capacity, as it can with Spot Instances. This is the most flexible option and the baseline that every discount is measured against.
• Reserved Instances (RIs). A billing discount in return for committing to a specific instance configuration (such as instance family, Region, operating system and tenancy) for a one-year or three-year term. You can pay all upfront, partially upfront or nothing upfront; the more you pay upfront, the larger the discount. An RI is not a running server. It is a discount that AWS applies automatically to matching usage.
• AWS Savings Plans. A commitment to spend a consistent amount on compute each hour, for a one-year or three-year term, in return for a discount. You commit to spend, not to particular instances, and usage above the commitment is charged at On-Demand rates. Compute Savings Plans apply to EC2 whatever the instance family, size, Region or operating system, and also to AWS Fargate and AWS Lambda. EC2 Instance Savings Plans are tied to one instance family in one Region and give a larger discount in return.
• Spot Instances. Spare EC2 capacity sold at a steep discount. AWS can reclaim a Spot Instance, after a short warning, when it needs the capacity back, so the work must tolerate interruption.
• Dedicated Hosts. A whole physical server for your use only. You see and control its sockets, cores and instance placement, which matters for software licences priced per socket or per core.
• Dedicated Instances. Instances on hardware dedicated to a single customer, while AWS decides which physical host they run on. You get single tenancy without managing hosts.
• On-Demand Capacity Reservations. Reserve EC2 capacity in a specific Availability Zone for as long as you need it, with no long-term commitment, and cancel at any time. You pay the On-Demand rate for the reserved capacity whether or not you use it, and you can combine it with Savings Plans or regional RIs to get a discount on top.

Two of these trade on different axes. Savings Plans and regional RIs reduce price but do not guarantee capacity; Capacity Reservations guarantee capacity but are not a discount.

Storage options and tiers

Storage is priced by how it is built and how often you expect to read it. The broad families, covered in lessons 18 and 19, are object storage (Amazon S3), block storage (Amazon EBS volumes attached to instances) and file storage (Amazon EFS and Amazon FSx).

Within Amazon S3, the storage classes form a ladder. Moving down it, storage gets cheaper, while reading the data gets slower, costs more per retrieval, or both:

• S3 Standard: frequently accessed data, millisecond access, stored across several Availability Zones.
• S3 Intelligent-Tiering: data with unknown or changing access patterns; S3 moves each object between access tiers for you.
• S3 Standard-Infrequent Access (S3 Standard-IA): data read rarely but needed in milliseconds, across several Availability Zones.
• S3 One Zone-Infrequent Access (S3 One Zone-IA): the same pattern, stored in one Availability Zone for a lower price, so suited to data you can recreate.
• S3 Glacier Instant Retrieval: archive data read about once a quarter that must still come back in milliseconds.
• S3 Glacier Flexible Retrieval: archives returned in minutes to hours.
• S3 Glacier Deep Archive: the lowest-cost class, for data kept for years and retrieved within hours.

S3 Express One Zone sits outside that ladder: a single-Availability-Zone class built for the lowest latency and highest request rates. It is priced for speed, not as a cheap tier.

When to use each compute purchasing option

Scenario questions usually describe the workload, and the workload picks the option:

The trap answers follow a pattern. Spot is cheapest, but it is wrong for anything that must not be interrupted. Reserved Instances save money, but they are wrong for a one-off job that cannot use a long term. A real fleet often mixes options: reservations or Savings Plans for the steady base load, On-Demand for peaks, and Spot for interruptible work.

Describing Reserved Instance flexibility

Reserved Instances come in two offering classes:

• Standard RIs give the largest RI discount. You can modify some details, such as the Availability Zone or the instance size within the same family, but you cannot exchange them for a different instance family. You can sell unused Standard RIs to other customers on the Reserved Instance Marketplace.
• Convertible RIs can be exchanged during the term for other Convertible RIs with a different instance family, operating system or tenancy. That flexibility costs you some of the discount.

RIs also have a scope:

• A regional RI applies its discount to matching usage in any Availability Zone in its Region, but reserves no capacity. A regional RI for Linux with default tenancy also has instance size flexibility: the discount applies across sizes in the same instance family, in proportion to size. If you resize to a larger instance in the same family, the RI still covers a share of the larger instance’s usage, and you pay On-Demand rates for the rest.
• A zonal RI applies to one named Availability Zone and also reserves capacity there, but loses the size flexibility.

An RI does not move between Regions, and an unused term is not refunded.

Describing Reserved Instance behavior in AWS Organizations

AWS Organizations, covered in lesson 24, can combine many AWS accounts under consolidated billing. For billing purposes, all the accounts in the organisation are treated as one.

By default, RI discounts are shared across every account in that consolidated billing family. If one member account bought RIs it no longer fully uses, and another account runs matching instances, the unused discount covers that other account’s usage automatically. The account that bought the reservation benefits first. Nobody has to move anything by hand. Savings Plans discounts are shared in the same way.

Sharing is a default, not a rule. The management account can turn off RI and Savings Plans discount sharing for chosen accounts in the billing preferences of the Billing and Cost Management console, so that their discounts stay inside the account that bought them. Moving an account into a different organisational unit does not do this; organisational units affect policy inheritance, not discount sharing.

Data transfer costs: in, out, between Regions and within a Region

Data transfer is the charge newcomers most often forget. The pattern to learn:

• Into AWS from the internet: generally free.
• Out of AWS to the internet: charged, per gigabyte, with lower rates as volume grows.
• From one AWS Region to another: charged, even though the data never leaves AWS.
• Between Availability Zones in the same Region: normally charged.
• Within one Availability Zone, using private IP addresses: generally free.
• Between Amazon S3 and EC2 in the same Region: generally free.

The design lesson follows from the list: keep chatty components in the same Region, keep large flows off the public internet where you can, and remember that a multi-Region or multi-AZ design buys resilience at the price of transfer charges. Amazon CloudFront can reduce what you pay to serve users, because transfer from AWS origins to CloudFront is not charged and CloudFront’s own delivery rates differ from direct transfer out.

How storage options and tiers are priced

Each storage service charges on different dimensions, and the exam expects you to know which ones.

Amazon S3 charges for:

• the amount stored, at a rate set by the storage class;
• requests, such as reading, writing and listing objects;
• data transfer out of the Region or to the internet;
• retrieval from the Infrequent Access and Glacier classes, which also carry minimum storage durations, so deleting an object early still incurs a charge.

S3 Intelligent-Tiering adds a small monitoring and automation charge per object but has no retrieval fees, which is why it suits unknown access patterns. Inbound transfer is generally free, the number of IAM users makes no difference, and general purpose buckets live in a Region, not in an Availability Zone.

Amazon EBS charges for the storage you provision on each volume, whether or not you fill it, and keeps charging while the volume exists, even if the instance is stopped. Some volume types also charge for performance you provision. EBS snapshots are charged for the data they store.

Amazon EFS charges for the data you actually store, with nothing to provision, and offers cheaper storage classes for files you read less often.

What the exam is really asking

Most pricing questions turn on one phrase. “Can be interrupted” points to Spot. “Steady for years” points to Reserved Instances or Savings Plans, and “may change family” to Convertible RIs or a Savings Plan. “Lambda and Fargate” points to Compute Savings Plans. “Per-core licence” points to Dedicated Hosts. “Guaranteed capacity, no commitment” points to Capacity Reservations. “Kept for years at the lowest cost, retrieval in hours is acceptable” points to S3 Glacier Deep Archive, and “rarely read, but in milliseconds” to S3 Glacier Instant Retrieval. And when a question asks what is free, the safe answer is data coming in.