Welcome to the third article in our "Mastering Cloud Cost Optimization" series. This series was designed to help cloud users maximize the value of the cloud by sharing best practices and expert knowledge based on our experience. In this article, we will focus on leveraging the right cost model for your cloud workloads.
If you missed the previous articles, you can access them from the below:
- Mastering Cloud Cost Optimization: The Principles
- Mastering Cloud Cost Optimization: Frameworks for Success
Choosing the right pricing model for your workload is one of the most common and effective methods of reducing costs on public clouds. To be successful in this effort, users must first understand their workloads and categorize them accordingly - for example, Production workloads, seasonal workloads, Dev/QA/Testing workloads, legacy applications, modern applications, etc. Then, users must apply the best cost model to the workloads to maximize the savings potential.
Leveraging the right pricing model is also one of the five cost optimization pillars of the AWS Well-Architected Framework for cost optimization:
Cloud Cost Models Overview
All major cloud providers offer the same three types of cost models when users consume:
- On-Demand / Pay-as-you-go – this model is the default costs users will be charged when consuming cloud resources. The model does not offer any discounts and users will be billed the full price (except for Google’s Sustained Use discount model) and do not require any commitment from the user. The usage of this model is measured by the second and billed monthly.
Ideal workloads and use cases for this model include short-term or seasonal workloads with expected uptime of hours, days, or weeks; Long-term workloads that do not have to run 24/7 and can be scheduled for suspension off hours.
- Reservations – the reservations model is based on the concept of users reserving capacity on their cloud vendors' infrastructure. Discounts depend on various variables such as compute type, location, reservation term (1-yr or 3-yrs.), and payment terms. Potential discounts can reach 70% and vary between clouds and reservation types.
Ideal workloads and use cases for this model include long-term workloads that are steady-state with their resources’ utilization and uptime, for example, web servers, databases, proxy servers, bastion hosts, etc.
- Spot / Preemptible – this is a unique model that offers the highest potential savings – users can see up to a 90% discount on AWS and Azure and up to 80% on GCP. The model is based on current available capacity within the cloud provider datacenter - the catch, however, is that if the cloud vendor needs the capacity back, the workloads that are using it must be terminated within a very short time window.
Ideal workloads and use cases for this model include modernized and fault-tolerant applications, batch jobs, media encoding and rendering, and Big Data processing. It is possible to mix spot instances with other models to reduce overall costs - for example, a Kubernetes or Big Data cluster with several nodes running on-demand (or utilizing reservations) and the rest of the cluster’s nodes using spot instances/VMs.
Cloud Discount Models Overview
In order to attract and retain users, cloud providers also offer various discount models that can be applied on top of the above cost models. The discount models vary from one cloud provider to another and require extensive negotiations to maximize discounts, but the main discount models are:
- Commitment – the commitment discount model is offered to large enterprises who wish to reduce costs by negotiating special rates for specific services based on their business needed. The discount is based upon a long-term period and committed volume. However, this model requires the organizations to ensure they consume the resources they committed to within the negotiated term to avoid loses. The main programs are Amazon Enterprise Discount Program (EDP) and Microsoft Azure Enterprise Agreement (EA).
- Bring Your Own License (BYOL) – the BYOL model is designed to allow users to bring and apply their pre-purchased software and operating systems licenses to workloads they deploy on the cloud if using one of the cloud vendors images - for example, images offered on their respective marketplace. It is important to note that there are several limitations and legal requirements when using licenses on certain cloud vendors, so special attention is needed when leveraging this model. For example, Azure Hybrid Benefits for SQL and Windows will eliminate associated licensing costs when deploying Windows and/or SQL workloads on Azure.
- Usage/Volume Discounts – this discount model offers a tier-based discount that is usually applied automatically. The model will apply graduate discounts as users consume certain cloud services - once they exceed a pre-defined threshold, a discount will be applied. Furthermore, it is possible to negotiate exclusive discounts with the cloud provider for certain services. Volume discounts can offer between 10%-15% discounts at the higher tiers. Examples of such programs include AWS Volume Discounts and Google Sustained Use. To maximize the potential saving of this model, it is recommended to consolidate all related cloud accounts under a single Organization / a billing account to achieve higher tiers of usage volumes.
There are also special offers and other pricing mechanisms that are noteworthy:
- Free Tier – almost all major cloud vendors offer a free tier to allow new users to experiment and test their cloud services, some up to 12 months (with some conditions and limitations). The main offerings are AWS Free Tier, Azure Free account, and Google Cloud Platform Free Tier.
- VM Promotion Pricing – occasionally Microsoft will announce special VM price promotions for specific SKUs. These promotions are time-limited and will offer substantial discounts. For example, Azure GPU and HPC VM Price Promotion, which was available between May and November 2019 and included a 56% discount for NCv1-series and 40% for Hv1 and NVv1-series.
- Dev & Education Pricing/Credits – several cloud vendors offer special discount and credit programs geared toward developers and education users (faculty, students and research centers). These discounts are designed to expose professionals (and professionals to be) to the cloud providers’ services without paying the full price. Examples include Google for Education and Azure Dev/Test Pricing.
Using the right cost model for your workloads should be the cornerstone of any cost optimization strategy. Although the leading cloud providers offer similar cost models types, provider offering has its unique benefits and nuances. As more organizations embrace multicloud approach, the need to master the differences between the models across clouds becomes increasingly critical. The cost models (and discounts) available on each cloud should be an important factor in the decision of where to deploy your next application.
Stay tuned for the next article in the series, ‘Mastering Cloud Cost Optimization: Increasing Elasticity with Scheduled Workload Suspension’ -- as its name implies, the article, by ParkMyCloud, a Turbonomic company, will focus on one of the most critical and impactful elements of cost optimization: scheduled resources suspension. Don’t miss it!