FinOps: Practical Strategies for Optimizing AWS Cloud Spend

Introduction

In the rapidly evolving landscape of cloud computing, financial operations (FinOps) has emerged as a critical discipline, blending financial acumen with operational expertise to drive cost efficiency and strategic business value. We are following our foundational article on “FinOps: Why, What, and How”, this piece delves deeper into practical, actionable strategies that organizations can employ to optimize their AWS cloud spend.

As businesses scale, so do their cloud expenses, often exponentially. The need to harness and control these costs cannot be overstated, as unchecked cloud spend can quickly erode the efficiencies and agility that cloud technologies are meant to offer. This article extends the principles discussed previously by focusing on advanced cost optimization techniques such as resource tagging, rightsizing, nightly shutdowns, and more.

Tagging Best Practices

Effective resource tagging is foundational to cost management in AWS. Proper tagging facilitates precise tracking and allocation of costs and enhances operational management across different organizational departments.

Why Tagging Matters

In AWS, every service or resource that incurs cost can be tagged. Tags are key-value pairs attached to resources, that can be used for many purposes. In our case, we use them for better filtering, like allowing organizations to attribute costs to the appropriate teams, projects, or environments.
With proper tagging, it becomes easier to discern the purpose and ownership of cloud spend, leading to potential budget overruns and decreased accountability.

Developing a Tagging Strategy

• Standardize Tags: Define a set of tag keys relevant across departments, such as CostCenter, Project, Environment, and Owner. This consistency simplifies reporting and cost allocation.
• Mandatory Tagging Policies: Implement policies requiring specific tags on all resources at creation, using AWS Organizations’ service control policies (SCPs).
• Automate Tagging: On resource creation by an IAC tool like Terraform, you can automate tagging. This will ensure error-free methodology and consistency.
• Regular Audits: Perform regular audits of resource tags to maintain accurate cost tracking.

Practical Applications

• Cost Allocation Reports: AWS cost allocation reports can break down spending by tags, providing detailed financial insights.
• Budget Monitoring: Set budget alerts for specific projects or departments using tags to prevent cost overruns.
• Operational Efficiency: Tags can manage compliance and security, enforcing policies based on the environment (Prod, Dev, Test).

Rightsizing Resources

Rightsizing ensures that cloud resources match the operational needs of an application, avoiding overprovisioning or underprovisioning. The Rightsizing is focused mainly on CPU and RAM.

General Methods

• Performance Monitoring: Use tools like AWS CloudWatch to track instance performance and adjust based on metrics.
• Load Testing: Simulate different load scenarios to understand resource usage patterns.
• Utilization Metrics: Establish benchmarks for resource usage to guide scaling decisions.
• Historical Data Analysis: Review historical performance data to predict future resource needs.

Leveraging Karpenter on EKS

One of the challenges in Kubernetes Right-Sizing is picking the correct machine type (x.medium, x.large, etc…) for our node groups.
This can be approached by using Karpenter.
Karpenter automates provisioning and scaling resources on a Kubernetes cluster based on actual needs.

• Event-Driven Scaling: Karpenter responds dynamically to real-time demands, reducing waste.
• Proactive Optimization: Continuously evaluates resource utilization and adjusts provisioning.
• Integration with AWS Services: Seamlessly integrates with AWS services like EC2 and EKS.
• Cost-Effective Scaling: Manages the resource lifecycle efficiently, reducing the total cost of ownership.

Implementing Nightly Shutdowns

Implementing nightly shutdowns for non-essential resources can significantly reduce AWS costs, especially in development and testing environments.

For instance, if we have a development machine, then it is up for 168 hours a week. If we turn it off at night (say… 21:00 to 08:00) and on weekends then the up and running hours are 65 hours a week. The payment for the storage still exists, but compute fees are reduced by ~60% !!!

This can be done on ec2, RDS, and in a Kubernetes cluster, we can reduce node number to 0.

Strategy

• Identify Eligible Instances: Determine which instances can be safely shut down during off-hours.
• Schedule Shutdowns: Use AWS Instance Scheduler to automate startup and shutdown processes.
• Communication and Enforcement: Inform stakeholders and implement policies for new non-production instances.
• Monitoring and Adjustments: Regularly monitor and adjust shutdown schedules as needed.

There are mature organizations that implement smarter processes. For example, all ec2 machines are shot down at 21:00. But to turn a machine on, the developer must implement a manual action.

Leveraging AWS Lambda

Use AWS Lambda for flexible control of shutdown strategies.

• Check for Idle Instances: Create functions that automatically shut down low-utilization instances.
• Use Tags for Exceptions: Implement tagging strategies to prevent auto-shutdown for certain instances.
• Log and Notify: Set up notifications to maintain transparency and manage expectations.

In general, organizations avoid this step because it requires more effort than a usual configuration. Doing a proper ROI is a good start to convince decision makers.

Cost Thresholds for AWS Lambda

AWS Lambda provides a cost-effective serverless computing service, but understanding its cost thresholds is crucial.

Understanding AWS Lambda Pricing

AWS Lambda pricing is based on the number of requests for your functions and the time your code executes, which is measured in gigabyte-seconds (GB-seconds). This measurement is a combination of the memory allocated to your function and the time it takes for your function to execute. The cost-effectiveness of Lambda generally increases as the granularity of the tasks it performs increases.

When Lambda is Cost-Effective

• Event-Driven Environments: Ideal for workloads requiring short bursts of processing.
• Variable Workloads: Suitable for applications with highly variable traffic.
• Microservices and APIs: Well-suited for microservices architecture and APIs.

When Lambda May Not Be Optimal

• Long-Running Functions: Functions with long execution times may be more cost-effective on EC2 instances.
• High Memory Requirements: Large memory functions might find EC2 instances cheaper.
• Persistent Connections: Workloads requiring persistent connections may not be suitable for Lambda.

Analyzing Break-even Points

• Benchmark Against EC2: Compare the cost of running similar workloads on EC2.
• Experimentation and Monitoring: Use AWS CloudWatch to track function performance and make informed decisions.
• Optimization Strategies: If Lambda remains the preferred option, consider optimizing your functions by reducing execution time and memory usage, which directly lowers costs. Techniques include fine-tuning your code, reducing the number of libraries, and using lighter dependencies.

Optimizing S3 Storage Costs

Amazon S3 (Simple Storage Service) is an essential component of AWS, offering scalable object storage for data of all sizes. While S3 pricing is competitive, costs can escalate with increased usage, especially without proper management and optimization strategies. This section explores effective methods to control and reduce expenses associated with S3.

Understanding S3 Pricing

S3 pricing is primarily determined by three factors:
1. The amount of data stored.
2. The number of requests (GET, PUT, DELETE, etc…).
3. Data transfer costs.

Additionally, the chosen storage class impacts cost, with each class designed to fit different use cases based on data access patterns and durability requirements.

Key Strategies for Cost Optimization

Rightsize the Storage Class: AWS offers several storage classes tailored for different needs, which include:

• S3 Standard: Best for frequently accessed data.
• S3 Intelligent-Tiering: Automatically moves data between two access tiers based on changing access patterns.
• S3 Standard-IA (Infrequent Access): Suitable for data that is less frequently accessed but requires rapid access when needed.
• S3 One Zone-IA: Similar to Standard-IA but stores data in a single Availability Zone.
• S3 Glacier and Glacier Deep Archive: For archiving data with retrieval times ranging from minutes to hours.

Assessing and aligning data storage to the most cost-effective class depending on access patterns can result in significant savings.

Implement Lifecycle Policies: Automate the transition of data to more cost-effective storage classes using lifecycle policies. For example, move data to Standard-IA after 30 days of no access, and to Glacier after 90 days. Lifecycle policies can also automate the deletion of obsolete or unnecessary data, which further reduces costs.

Monitor and Analyze Storage Usage: Regularly review storage usage with tools like AWS Cost Explorer and S3 Analytics. These tools help identify cost drivers and inefficiencies, such as old data that can be archived or deleted.

Optimize Data Transfers: Data transfer costs can be minimized by keeping data in regions closer to your users and by leveraging AWS’s CloudFront for content delivery. Additionally, consider using S3 Transfer Acceleration for faster uploads across long distances.

Delete Unnecessary Data: Regular audits to identify and delete outdated, redundant, or unnecessary data can lead to direct cost savings. Automated scripts or manual processes can be employed to clean up old buckets and objects that are no longer needed.

Use Requester Pays: For buckets where data is accessed by external parties, consider enabling the ‘Requester Pays’ option. This shifts the cost of data retrieval and request operations to the requester rather than the bucket owner.

Leveraging Reserved Instances and Savings Plans

AWS offers financial instruments like Reserved Instances (RIs) and Savings Plans to reduce costs.

These instruments are based on “commit to 1 or 3 years, and get a discount”.

Reserved Instances

If you can commit to 1 or 3 years for an ec2 of a specific family/type and specific AZ, then you can pay in advance (all payments in advance, partial pay in advance plus monthly fees, and all in monthly fees).
The discount is calculated based on the above parameters.
E.g. committing to 3 years instead of 1 will have a better discount.
Notice: You are not committing to a specific ec2, but to any ec2 that meets the above criteria. So, if you buy one RI, it looks for a matched ec2. Once found, the RI is attached to it, and you can get the discount. If this ec2 is terminated, another matched ec2 will be attached.

• Assess Usage Patterns: Analyze instance usage to identify candidates for RIs.
• Choose the Right RI: Select Standard or Convertible RIs based on flexibility needs.
• Manage RIs Actively: Regularly review and adjust RIs to maximize ROI.

AWS Savings Plans

The methodology of AWS Savings Plans involves committing to a consistent amount of usage, measured in dollars per hour, over a one- or three-year term. In exchange for this commitment, you receive lower prices on your AWS compute usage, including Amazon EC2, AWS Fargate, and AWS Lambda, regardless of instance type, region, or operating system. This flexibility allows you to change your usage patterns and still benefit from the savings, offering a more adaptable and cost-effective solution compared to Reserved Instances.

• Evaluate Coverage: Determine if Savings Plans cover your typical usage.
• Balance Commitment and Savings: Set spending levels based on past usage trends.
• Monitor and Adjust: Continuously monitor Savings Plan performance against actual usage.

Conclusion

Mastering AWS cost optimization requires a blend of technical strategies, financial instruments, and continuous monitoring. By implementing these practices, organizations can ensure their cloud infrastructure is cost-effective and capable of adapting to evolving business requirements. The ultimate goal is to foster a culture of cost awareness and efficiency, driving both fiscal prudence and technological innovation.