The Problem With (Traditional) Reserved Instances
Frederik Bussler

Frederik Bussler

May 17, 2022 · 4 min read

https://s3.amazonaws.com/usage-event-images/1652814391137-lock.png?X-Amz-Algorithm=AWS4-HMAC-SHA256&X-Amz-Date=20220625T220104Z&X-Amz-SignedHeaders=host&X-Amz-Expires=172800&X-Amz-Credential=AKIA4AXCT4E7GLDVACC4%2F20220625%2Fus-east-1%2Fs3%2Faws4_request&X-Amz-Signature=bd628724a3c6527fcd44b2a1b99aec291f203b114e48d466fbbdf649edf3e28d

Like any SaaS business, cloud providers struggle with customer churn. When it comes to AWS, one of the key ways they combat this is by offering Reserved Instances (RIs).

RIs are effectively locked-in subscriptions for AWS services. You commit to using a certain amount of resources for a set period of time, and in return you get a discount on your bill. This sounds like a great deal for both parties – AWS gets guaranteed revenue, and customers get lower costs. Beyond AWS, other cloud providers have similar offerings. For example, Azure's equivalent is called Azure Reserved Virtual Machine Instances.

Let’s look at some of the issues with traditional RIs, particularly upfront RIs, and how Usage.AI solves these issues with no-upfront RIs that are bought and sold in real-time as your needs change.

Misaligned Incentives

Cloud spending is expected to reach a tremendous $1.3 trillion by 2025, from an annual growth rate of almost 17%, according to IDC. For cloud providers to achieve this enormous growth, they need to keep customers locked in.

By locking customers into a long-term contract, providers can reduce the likelihood of that customer switching to a competitor. Once you’ve committed to an RI, it's in their best interest to keep you locked in, even if there are better options available. Therefore, the biggest problem with RIs is that they create misaligned incentives between the cloud provider and its customers.

Lack of Flexibility

Another big problem with RIs is that they lack flexibility. This is most apparent when it comes to changes in usage patterns.

For example, let’s say you have a workload that's variable in nature. Some months it may use more resources, and some months it may use less. With an RI, you're effectively paying for capacity whether you use it or not. This can lead to wasted spend, as you're paying for resources that you don't necessarily need.

Additionally, RIs are often non-refundable. So if your plans change and you no longer need the capacity, you're still on the hook for the entire duration of the contract.

This lack of flexibility can be a major problem for customers who find themselves in a situation where their usage patterns have changed and they can no longer utilize their RIs.

Hard to Forecast Usage

Prediction is hard, especially when it comes to cloud usage. There are a lot of factors that can affect resource utilization, from CPU and memory usage to disk I/O and network throughput. Moreover, given the dynamic nature of cloud workloads, it can be difficult to get an accurate forecast.

With many instance types and little visibility into cloud bills, businesses can often find themselves unexpectedly spending more on Reserved Instances than they thought. For example, a company might miscalculate its traffic patterns and over-provision instances, leading to wasted spend.

Compounding the problem, properly tagging cloud resources is often difficult. This makes it hard to attribute cost to the right department or application. Finance and engineering teams might have different priorities when it comes to Reserved Instances, which can further complicate decision-making.

Complicated Pricing Structure

Lastly, the pricing structure for RIs can be complicated and difficult to understand. There are numerous options and variables to consider, which can make it hard to know if you're getting a good deal.

To add to the confusion, AWS frequently introduces new RI offerings and discounts. This means that customers who sign up for an RI today may not be getting the same deal as those who signed up a month ago. 

Even more confusing is the fact that some RIs, or Convertible RIs, can be modified after they've been purchased, while others cannot. This again makes it hard to know if you're getting a good deal.

All of this complexity makes it very difficult for customers to fully understand and take advantage of Reserved Instances. As a result, many customers end up overpaying for their RIs, or not getting the full benefit of them.

Complexity in Instances

Beyond the RI options and pricing, the instances themselves can be complicated. AWS offers over 350 Amazon EC2 instances, which can be further customized with a variety of options.

This wide range of choices is great for customers who need a specific instance type to meet their needs. However, it also adds to the complexity of choosing an RI. Customers need to carefully consider their instance type, size, and options to make sure they're getting the right RI for their needs.

Additionally, customers need to be aware that their instance needs may change over time. As business needs change, so too will the instances that are needed to support them. This means that customers who don't monitor their usage and adjust their RIs accordingly may end up paying more than they need to.

Real Consequences: Pinterest Loses $20 Million

The complexity of RIs can have real consequences for businesses that use them. For example, Pinterest signed up for an RI with AWS in order to get a discount on their usage. However, due to changes in their business, they ended up needing more capacity than they had initially reserved.

This resulted in Pinterest having to buy additional capacity at a higher price. This is because cloud providers can control the prices of additional capacity, and often charge significantly more than the rates they offer for RIs.

In the end, Pinterest ended up spending $20 million more on AWS than they had originally expected. This is a direct result of the complexity of RIs and the difficulty of understanding how they work.

Pinterest's losses are just a drip in the bucket compared to the $500 billion in lost market value that the cloud has wiped off the books of major public companies. But it's a big reminder that even the savviest technology users can get tripped up by the nuances of cloud pricing.

Publicly-listed companies are well-equipped to weather a $20 million charge. But for small businesses and startups, unexpected costs like this can be devastating. Usage.AI recently wrote a Forbes article on how cloud costs can even bankrupt SMEs, highlighting examples of statups facing sudden $72,000 bills, all the way to large firms leaving the cloud entirely.

The cloud has been a boon for businesses of all sizes. But as these examples show, it's important to understand the potential risks and costs associated with using Reserved Instances. Otherwise, you may end up with a nasty surprise.

The Solution: An Automated RI Marketplace

Usage AI has created a solution to this problem – an automated RI marketplace.

By combining automation technologies and the EC2 Reserved Instance marketplace, Usage AI is able to buy and sell RIs on behalf of customers. This means that customers can cancel their RIs at any time, without penalty. And if they need more resources than their RI allows, they can simply purchase additional resources on the marketplace.

Usage AI takes care of the entire process – from finding buyers for customers' unused RIs, to automatically buying and selling RIs on the marketplace. This gives customers the flexibility they need, while still getting the discount associated with an RI. Moreover, Usage AI uses no-upfront RIs, so customers don't have to pay anything upfront, and you only pay for 20% of what you save.


The Author

Frederik BusslerContent Marketer
Frederik Bussler
Content Marketer

Linkedin

Twitter

Frederik is a content marketing consultant with experience across startup, mid-market, and enterprise companies, helping them to develop and execute long-term content strategies.

Tools

EC2 CalculatorEC2 Pricing
Powering the next generation of cloud-intelligent software.
Free Consultation

usage logo
PrivacyTerms

Copyright Ⓒ 2022 Usage AI.