Learn how FinOps teams can optimize AWS costs by following proven best practices and tactics.
Cutting down an AWS bill starts with understanding the services and resources you’re paying for. The next step is the actual optimization process. The goal is to optimize AWS expenses without compromising performance, reliability, and security.
Here are six FinOps and DevOps best practices for AWS cost optimization to help you achieve that.
1. Start with cost visibility
There’s only so much digging you can do by looking at your bill. The AWS Cost Management console is a feature integrated into the AWS billing. When you use these tools together, you can view your costs and billing holistically.
The console includes several free-to-use features that help you better understand, organize, and detect anomalies in your cloud billing. Here’s a selection:
- AWS Cost Explorer
- AWS Budgets
- Cost Anomaly Detection
AWS Cost Explorer
Cost Explorer visualizes cost data for a deeper analysis. You can filter graphs to include cost allocation tags, availability zones, regions, and other purchase and service options. You can also forecast future expenses based on past data to better understand your spending and optimize your cloud costs.
AWS budgets
Setting budgets in AWS helps ensure you don’t exceed your forecast. Or, at the very least, you’ll receive a notification so you can take immediate action to prevent your spending from spiraling out of control.
The tool uses AWS Cost Explorer to show a visual representation of your budgets, including forecasts for estimated costs, to help you better visualize your spending. You can also set notifications to alert your team when you reach certain budget levels – allowing you to stay on track without any nasty surprises.
Cost Anomaly Detection
The Cost Anomaly Detection tool alerts you when a potential irregularity is detected (for example, a rapid cost increase). This allows you to quickly identify the problem, keeping your cloud spending on budget without any unnecessary hiccups.
Cost reporting is essential, but it’s just the first step
Cost reporting is incredibly important, but it won’t reduce your costs on its one. You need more than cost reports to make a real difference in your setup. At best, what monitoring and reporting tools give you are static recommendations that require manual implementation.
If you’re running a small deployment, a human engineer might be able to make it work. At least for a little while, until you start scaling.
Applying all the recommendations manually and regularly takes time, which costs money – not to mention all the lost ad-hoc optimization opportunities during sudden traffic spikes.
Kubernetes cost optimization
Monitor organization-wide and cluster-level resource spending. Automate resource allocation and scale instantly with zero downtime. Learn more
2. Choose the right compute resources
AWS offers more than 700 instance types and sizes today across almost 40 regions. Moreover, it offers variable performance for similar instance types. A more expensive instance doesn’t always translate into better performance.
So, how do you pick the best compute instance for your workload?
Here are a few steps:
Create your base requirements
Ensure that all computational aspects are covered, including memory, SSD, network connectivity, CPU (architecture, count, and processor type), and memory.
Choose the right instance type
A variety of CPU, memory, storage, and networking configurations are available, each packaged in instance types tailored for specific capabilities (for example, machine learning workloads such as model training or inference).
Configure your instance size
Keep in mind that the instance needs to be large enough to support your workload and have options like bursting available if needed.
Analyze various pricing models
AWS offers various pricing models: on-demand (pay-as-you-go), reservations, Savings Plans, Spot Instances, and dedicated hosts. Each of them comes with unique offers, benefits, and drawbacks.
3. Scale resources to match real-time demand
Most cloud-based applications experience usage changes, but balancing costs and performance remains a challenge.
If you leave your tab open, traffic spikes can result in a large, unexpected cloud bill. If you place stringent limits on your app’s resources, a large number of new visitors can cause your app to fail.
There are cloud cost management systems that monitor your usage and instantly notify you if it exceeds predefined limits or shows unusual patterns. These tools might offer helpful suggestions for modifying your cloud resources to meet your current needs.
However, increasing your cloud’s capacity manually takes effort and time. In addition to monitoring everything that occurs within the system, you typically have to:
- Adapt quickly to fluctuations in traffic volume and resource usage for every virtual machine across all of your services
- Make sure that modifications made to one workload don’t negatively impact other workloads
- Create and oversee autoscaling groups to ensure they are configured with the appropriate resources for your tasks.
Autoscaling to the rescue
All of these tasks can be automatically completed via autoscaling, which also lowers cloud expenses.
Autoscaling is an excellent approach for optimizing cloud costs when used with dynamic cloud-native systems such as Kubernetes. Your optimization efforts will be more impactful, and there will be less waste in running the application, the tighter your scaling methods are designed.
Real-time autoscaling solutions use business metrics to determine the optimal number of instances. If there is no more work to be done, they can scale up, down, or to zero. Such systems ensure that the resources used continuously meet the application’s needs.
Check out this graph to see how requested vs. provisioned resources differed and how that difference shrank as soon as the team utilized the Cast AI autoscaling feature (marked by “Start of optimization”):
4. Remove orphaned resources
The cloud makes it simple to start an instance and then forget to shut it down. Because of this, many teams deal with orphaned instances, that is, instances that don’t belong to anyone but still end up on your AWS bill.
This issue is especially severe in large organizations where there is no centralized visibility into resources and multiple programs are underway simultaneously.
How can you find and shut down unnecessary instances? This is another situation where automated AWS cost optimization proves useful.
Automated cloud optimization solutions can continuously monitor your consumption for inefficiencies and reduce resources whenever it makes sense. To save your cloud expenses, they can also terminate inactive instances and processes.
5. Increase savings by using Spot Instances
Buying idle capacity from Amazon Web Services makes sense because Spot Instances cost up to 90% less than on-demand instances. There’s a catch, though: the vendor can take these resources back at any time.
Before joining the Spot club, ensure that your workloads are Spot-ready:
Assess your workload
Can your workload tolerate disruptions? How much time will it take to finish the task? Is this workload essential to your application? Questions such as these, along with others, help qualify a workload for Spot Instances.
Review the AWS offering
Examining less well-known compute instances is a good idea because they are less likely to experience interruptions and have longer operating lifespans. Always check an instance’s interruption frequency before choosing one.
Place a Spot Instance bid
Decide on the maximum price you’re willing to spend for the Spot Instance of your choice. Keep in mind that it will only last as long as the market price meets your offer. The general rule in this case is to set the maximum price at the on-demand level.
Group Spot Instances
Group your Spot Instances to request multiple instance types simultaneously, improving your chances of snatching an instance.
Expect to invest a lot of time in the configuration, setup, and maintenance tasks above. That is, unless you choose to automate Spot Instance provisioning as Yotpo did.
Yotpo automates Spot Instances, cuts 40% in cloud costs and saves time
6. Choose the right solution to manage AWS costs
To take charge of their cloud expenses, companies use a combination of cloud cost optimization tools:
- Cost visibility and allocation – You can determine the source of your spending by using a range of cost allocation, monitoring, and reporting tools. Since real-time expense monitoring promptly notifies you when you surpass a certain threshold, it’s helpful in this situation.
- Cost forecasting and budgeting – If you’ve crunched enough historical data and have a good sense of your future requirements, you can plan your budget and predict the number of resources your teams will need. Sounds easy enough? Cloud financial management is riddled with challenges.
- Legacy AWS cost optimization solutions: To get a full view of your cloud spend and identify possible areas for improvement, you will need to compile the data from the above. Numerous products on the market can help with that, but most of the time, they offer only static suggestions that engineers must manually execute.
- Automated, cloud-native cost optimization: Even if you have been carrying out excellent manual optimization, automation brings even better results without requiring more work from teams. The ideal strategy in this case is a completely autonomous, fully managed system that can respond rapidly to shifts in resource demand or cost-effectively.
Wrap up
AWS cost optimization isn’t a one-time exercise but a continuous effort. You need to constantly monitor our cloud resource usage, align it with your bill, forecast future requirements, and provision resources that deliver the best cost-performance ratio.
Or you can have an automation solution do that for you.
If you’re interested in exploring not only AWS cost optimization but also EKS-specific solutions, check out how we used our technology to reduce our Amazon EKS costs by 66% for one of our clusters: 8 Tips for Amazon EKS Cost Optimization.
