How to lower EC2 Costs

This article is meant for CEOs, CTOs, and Managers. The goal is lowering your AWS costs by up to 72 percent within a few minutes (depending on the instance type and plan), by answering a common question clients have: What options do I have at my disposal to lower the AWS bill?

Start with Savings Plans

Since I first wrote this, AWS added Savings Plans. Now, AWS actually recommends them over Reserved Instances. The concept is straightforward: instead of locking in a specific instance type, you commit to a set spend per hour, let's say 10 bucks, over one or three years. AWS then discounts everything that falls under that, up to 72% off on-demand prices.

There are two types. Compute Savings Plans cover any instance family, size, OS, tenancy, and Region. They also include Fargate and Lambda. Max flexibility but slightly smaller discounts. EC2 Instance Savings Plans are limited to one instance family in one Region but offer bigger savings, similar to a standard Reserved Instance.

Cost Explorer will suggest a good commitment level based on your actual usage, start there.

But why bother with Reserved Instances? They still matter if you need capacity reservation in a specific AZ or want to resell unused commitments on the marketplace. Savings Plans can't do those two things.

What are Reserved Instances?

If you commit to an instance for a period of either one or three years - with an upfront payment, a partial one, or none at all - then your price for an instance can be up to 72 percent lower. The beauty of this is that if you buy a Reserved Instance, you don't have to make changes to your instances - no restart or reconfiguration is necessary. This is because a Reserved Instance is a billing concept. If you have two instances and you buy one Reserved Instance, AWS will automatically apply the discount for the instance that matches the type you bought. The entire process is seamless for you.

What else should I know?

Standard vs. Convertible

There are two types of Reserved Instances: Standard and Convertible. Standard is suitable for a situation when you have an organized environment and you know what you need. Convertible (a bit more expensive) is for situations with a more dynamic environment that requires constant changes to your operating systems and instance families as well as types.

Normalization Factor

AWS gives you quite a bit of flexibility with the Normalization Factor approach. Each instance type has the following factors:

• nano: 0.25
• micro: 0.5
• small: 1
• medium: 2
• large: 4
• xlarge: 8
• 2xlarge: 16
• 4xlarge: 32
• 8xlarge: 64
• 10xlarge: 80
• 16xlarge: 128
• 32xlarge: 256

The table keeps growing as AWS adds sizes - it now goes all the way to 112xlarge with a factor of 896 - but the math stays the same.

This means that if you buy one large type, you can actually have four small instance types, or two small ones and four nano types. You can mix this as long as the total factor number matches the size of what you bought.

Heads up: you only get size flexibility with regional Reserved Instances, not zonal ones. And it only works for Linux or Unix with default tenancy. If you're running Windows, RHEL, or Suse, you have to stick to the exact size you bought to get the discount.

AWS also allows you to scale above what you have. For example, let's say you bought a large type for the year and you scaled up to the 2xlarge type for one week. AWS will charge you the On-Demand price, but only for the scaling factor that goes beyond what you paid for. They do the calculation automatically so you don't have to worry about it.

Be organized

The Scaling Factor is flexible, but best practice is to have a homogeneous instance type. This makes it much easier to manage and keeps costs to a minimum.

You can resell what you don't use

If you buy a one-year or three-year lease, and after some time you find that you don't need it anymore, or you've upgraded to a new type because of traffic to your site, then you can buy the new type and put the old lease on the market for bidding. While you won't be able to get 100 percent of the money back from the remainder of your old type, you'll still get a decent amount.

One more thing: only Standard Reserved Instances can be sold on the marketplace, not Convertible ones. Also, the reservation needs to have been active in your account for at least 30 days and have at least one month remaining on the term.

The roll-over strategy

Even with the flexibility AWS gives you, you may still be left with some unused instance types. Before you resell them, you should consider the following strategy: roll your instance types to the next-lower traffic site. For example, let's assume that you have a site with two parts, the front end and the back end. The back is the API, and the front delivers the site build using Angular, with some server rendering for good SEO.

Both projects run using a load balancer, and each LB has a minimum of two instances of type t3.small. You buy one type, a t3.large server, for one year and pay 100 percent upfront. You then use the Normalization Factor to fragment the one large into four small types.

Six months later you get so much traffic that you have to upgrade the back end to a larger type. In this case, you could change the instance type by buying a one-year lease for a large type and split that into two mediums.

Now you have two units to spare. You could add two more small (or one medium) instance types to your front end to handle the additional traffic.

The reasoning here is that more traffic will increase the load on both projects, but the front end will have the smallest load, and that can be handled by four small server types.

When the lease for the old types expires, you can switch the front end to two bigger server types or extend the lease for what you already have. AWS's flexibility means you have options to role-over instance types and save money.

To sum it up

Stop using On-Demand servers for long-term work. Use them only for quick tests, maintenance, or temporary auto scaling to handle spikes in traffic during the day. For continuous work, commit - a Savings Plan for most cases, Reserved Instances when you need the capacity guarantee - and spend your money elsewhere.

If you want a second pair of eyes on your account before committing to anything, I offer a fixed-price AWS audit that covers costs, security, and architecture in one report.