Decoding AWS’s complex, cost-cutting product—Reserved Instances

By
Mathew Pregasen
August 24, 2022

A friend of mine once quipped that Amazon must’ve retained a Chief Jargon Officer in the early days of AWS—Bezos’s infrastructure titan isn’t exactly known for descriptive product names. Between EC2, Redshift, Route 53, and dozens of other polygon-logo’d services, AWS is teeming with coded names behind industry-leading products. Today, we are going to discuss one of the most complex but crucial AWS offerings—Reserved Instances.

Before reading further, you should know three things about Reserved Instances (RIs).

  1. The Premise. RIs involve exchanging capital and freedom in return for long-term savings.
  2. The Form-Factor. Reserved Instances are not real instances or special servers. Reserved Instances is the AWS-equivalent of a lease. A billing construct. Nothing more, nothing technical. If you purchase a Reserved Instance, your servers, IP addresses, geographies entirely remain the same. Your engineers shouldn’t have to change anything.
  3. The Constraint. The final thing is that certain Reserved Instances can be altered mid-contract to include additional compute power.

Reserved Instances are a major piece of AWS Billing, accounting for $100B in revenue along side Saving Plans (SPs). We will explore how RIs are categorized, are purchased, and some additional services your company can trust to make RIs less befuddling.

Do I need an RI?

If AWS spend isn’t a major priority or you don’t foresee major upcoming changes to your infrastructure, considering RIs may not be worth your time. However, if your AWS spend is beginning to impact your margins or limit headcount, or, you are looking to cut spend across departments in general, exploring RIs could net massive 20-70% savings. You read that right—seventy percent.

Types of Reserved Instances

First, we need to understand what is available on the market. Amazon’s Reserved Instances vary across three dimensions: (i) contract length, (ii) type, and (iii) payment terms. Like any other good or service—except perhaps McDonalds chicken nuggets—the value of RIs is a function of how much you’re willing to pay (up front) and how much constraint you're comfortable with.

Let's expand all three dimensions.

Contract Length

You have two options if you buy an RI directly from AWS—1 year and 3 year commitments. I say directly because you can purchase partially used, often-discounted RIs on AWS’s RI marketplace (more on that below).

Payment Terms

AURI vs PURI vs NURI.png

For payments terms, there’s a new set of jargon you need to navigate—AURIs, PURIs, NURIs, which stand for All Upfront RIs, Partial Upfront RIs, and No Upfront RIs respectively. If we were to compare RIs to mortgages:

  • An AURI is akin to purchasing a house in cash. Stumping lump-sum payment, no interest fees.
  • A PURI, the middle option, equates to a variable down payment along side monthly payments for the remainder.
  • A NURI, which renders lessened yet still substantial discount, allows you to continue to pay your EC2 AWS bill on a monthly basis.

AURIs are sensible for businesses that have predictable server allotments for the next 1-3 years and in a position to take on the CapEx to reduce long term costs. NURIs appeal to cash-strapped startups, especially if they can't confidently predict their workloads over the next year.

Type

Standard RIs, Convertible RIs, and Scheduled RIs are three types of RIs.

  • Standard RIs are what you likely assume RIs to be—a commitment to use a resource for a lengthy period in return for a discount. This is best for businesses that have a resource they’ll predictable need without foreseeable upgrades.
  • Convertible RIs are similar to an iPhone Buy-Back program—you can exchange an RI whenever for an equivalent or more expensive RI. Convertible RIs are ideal for businesses expecting growth and not shrinkage.
  • Scheduled RIs are the most complicated—best for businesses that run servers on a schedule. In other words, Scheduled RIs are a more niche contract for businesses with backend services that turn on and off on a routine.

Of the these three, Standard RIs offer larger discounts over Convertible RIs. Notably, when I chatted with Usage.ai’s CEO, Kaveh Khorram, he expressed that Standard RIs are more popular than Convertible RIs, and most companies that explore Convertible RIs end up utilizing a Saving Plan (SP) instead, a separate cost-slashing product by AWS.

The Extremes

Reserved Instance Dimensions.png

A good way of understanding the range of RI savings would be by examining the extremes—the most-constraining versus the least constraining.

The best possible RI would be a Standard 3-Year AURI. That is, an RI that (i) cannot be upgraded, (ii) is for 3 entire years, and (iii) is paid up front. This would net you close to a hefty ~70% discount on your EC2 spend over a 3-year horizon. The other extreme would be a Convertible 1-Year NURI, which amounts to ~20-30% of savings.

Am I doomed if I want to get out of my contract?

Your gut should say yes, shouldn’t it? After all, that’s the whole point of RIs—you’re selling your soul—cough, sorry, your server flexibility—in return for a discount. But surprisingly, no, you aren’t doomed if you want out of your contract. You just need to find a buyer.

AWS has an RI marketplace. That’s right. Businesses can sell RIs they’ve purchased, setting their own price for the remainder of the contract.

AWS’s Reserved Instance Marketplace

AWS’s Reserved Instance Marketplace

Now, this isn’t a new opportunity to found an arbitrage fund. AWS exacts a 12% charge on any transaction, presumably to prevent any buy-sell chaos. Regardless, businesses may want to dump a RI they no longer need. The purchaser acquires the RI with time already spent, meaning shorter commitments with long-term equivalent savings.

Of course, finding a business that (a) knows about RIs, (b) is in need of that specific architecture, and (c) cares about the lessened term may be tough. Accordingly, RIs on the marketplace are often being sold at a loss, still a better outcome than paying for unnecessary compute. And, the prior savings of an RI will often offset that loss; RIs typically are a net positive for the customer’s wallet.

Of course, this means that surfing the marketplace is the best place to start when considering RIs. You just need to get lucky with finding the right architecture that matches your own stack.

This is still so confusing

It is. RIs are like any other AWS service: customers demand all the knobs and levers and that necessarily accrues complexity. But remember, RIs can massively impact your company’s long-term server spend, scoring discounts as high as 70%.

AWS’s Cost Explorer

AWS’s Cost Explorer

The best place to start is AWS’s built in Cost Explorer. You can discover Reserved Instance recommendations based on your existing stack and triage savings over the previously discussed dimensions.

Additionally, there are a few products on the market that help companies decide and purchase RIs. One is Usage.ai—they underwrite an RI for your business, netting you long-term savings without any commitment whatsoever. While Usage charges 20% of your savings, it will handle the purchase and re-sale of an RI if your business needs to alter infrastructures. Another is Vantage.sh, cloud-spend visualization platform, which helps you plan and purchase RIs for just 5% of the savings, but doesn’t underwrite the contract itself.

Closing Thoughts

Reserved Instances can render massive savings to your AWS bill. Given the range of RIs across commitment terms, instance types, and payment terms, there is likely a deal for you if you have some reasonable measure of predictability in your server stack.

Recommended Posts