Instance pricing models are highly confusing, and yet they're worth the investment required to learn more about them. For IT managers and CIOs, understanding cloud instance pricing models will go a long way to better visualizing cloud strategy goals and making roadmap discussions more productive.
All the major infrastructure-as-a-service (IaaS) service providers, including Amazon Web Services Inc. , Microsoft Azure and Google Cloud , offer pre-packaged server instance types. Variables within these instances include system-related components such as the type and number of CPUs, amount of memory, and storage sizes and types. Administrators select from these instance types based on system requirements of the application, database or service. Pricing for an instance can also vary in other ways, beyond virtual machine requirements. These variables include options regarding how one plans to pay and how available the instance needs to be. Paying an instance rate on an hour-by-hour basis will end up costing more compared to signing a multi-year contract for the same amount of server resources. Where things get cloudy, however, is this whole concept of instance availability. (See Amazon Bigger in IaaS Cloud Than Microsoft, Google & IBM Combined.)
Availability on a per-instance basis revolves around the idea that whatever application or service is running on the instance does not need access to those resources every second of the day. Therefore, to save money, cloud service providers essentially allow the sharing of the same hardware resources across multiple instances and multiple customers. Availability is important to cloud budget discussions because if done correctly, it allows IT administrators to appropriately right-size their virtual machines, but with the opportunity to gain discounted pricing by only making it available only when required. (See Wasting Money on Cloud: 6 Ways to Stop .)
Once IT infrastructure and application administrators choose the virtual machine instance options they require, the next decisions are choosing how the company will pay for said instance and how often it realistically needs to be available. This is where a tremendous amount of money can be saved, if done properly. For example, AWS "On-demand" instances are billed on an hourly basis. This is a great option for developers that only require instances be up for a short period of time. On the other hand, if administrators already know that specific instances are going to be running for months or years, AWS offers "Reserved" instances that require the signing of a long-term contract -- yet end up costing less compared to On-demand.
Availability requirements can vary from 24/7 to scheduled timeframes to whenever resources are free. Cloud providers offer the ability for discounts if you agree to essentially share server resources with others. AWS uses the term "Spot" instance. Azure and Google Cloud use the term "Low-Priority VM" and "Preemptible VM" when describing the same server-sharing concept. The idea is that the instance can be run at a far lower price compared to "always-available" instances if the resources are not required all the time. A perfect example of an IT service that would fit this bill would be applications that use batch processing, which can be run whenever instance resources are available.
Many find that the somewhat random access to server resources found in the lowest-cost availability instance types is not sufficient. Fortunately, cloud providers offer a pricing model that sits between always-available and randomly-available options. In the AWS world, the instance availability model is known as a "Scheduled" instance. Google Cloud and Azure use a slightly different way to schedule reserved instances for increased discounting. Regardless of the method, the outcome is the same. It allows administrators to schedule dates/times when the instance is required. The rest of the time, the instance is effectively turned off and made available to others. While this option provides tremendous flexibility while also saving money, keep in mind that these types of scheduled instances almost always require a 12-month or longer contract commitment.
For infrastructure IT professionals who work on cloud computing architectures, cloud instance types and design concepts are relatively trivial. However, IT managers and CIOs require a high-level understanding into infrastructure concepts when engaged in conversations related to the budget, capabilities and long-term strategies of cloud resources. Instance pricing models are easily one of the most confusing aspects of a cloud strategy. (See Cisco Looks to Allay Cloud Bill Shock.)
Cloud computing is going to be around for a long time. Therefore, as IT leaders, it's your duty to figure out instance pricing models -- and how best use them to your company's advantage.
If you can master the key differences between on-demand vs. spot vs. reserved instances, you're far less likely to get confused and go running to your favorite search engine in a quest for answers.
- Amazon Closes Cloud Performance Gap With 'Bare Metal Instances'
- AWS Draws Disney in as Cloud Customer
- Google Cloud Offering 'Preemptible' GPUs Plus Price Cut