Spot Instance Adoption: Lessons from 6 Case Studies

Spot instances can slash cloud computing costs by up to 90%, but they come with challenges like interruptions and pricing volatility. Businesses that design interruption-tolerant systems, automate processes, and monitor usage effectively can achieve massive savings while maintaining performance.

Key takeaways:

• Spot instances are discounted, surplus cloud resources, ideal for non-critical, fault-tolerant tasks.
• Challenges include sudden interruptions (e.g., 2-minute warnings) and fluctuating prices.
• Solutions involve robust architecture (e.g., checkpointing), automation, and diversification across instance types and zones.

Case studies show success stories:

• NFL saved £1.6M per season using 4,000 spot instances.
• Freshworks cut costs by 65% after overcoming automation hurdles.
• Amaysim reduced compute costs by 75% through architectural improvements.

For optimal results, combine resilience, automation, and continuous monitoring. Spot instances work best for flexible workloads, while critical tasks may still require on-demand or reserved instances. A hybrid strategy can balance cost and reliability.

Common Challenges in Spot Instance Adoption

While the cost savings offered by spot instances are undeniably appealing, they come with their own set of challenges that can complicate their adoption. Moving from theoretical cost analysis to practical implementation often uncovers hurdles that require careful planning. Let’s explore some of the key challenges, starting with the issue of unexpected interruptions.

Interruption Risks

One of the most significant challenges with spot instances is their unpredictability. Cloud providers can reclaim these instances with little notice when they need additional capacity for higher-priority tasks. Even if interruptions are infrequent, systems that aren't prepared for them can face serious disruptions.

For example, AWS provides a two-minute warning before reclaiming an instance, while Google Cloud and Azure often offer even less time. This leaves minimal opportunity to shut down processes gracefully, save data, or migrate workloads. To mitigate this, applications must be designed to handle interruptions from the beginning. This includes implementing robust checkpointing, ensuring that data persistence doesn’t rely on local storage, and creating systems that can restart seamlessly. Long-running batch jobs are particularly vulnerable, as interruptions can result in wasted processing time.

Persistent storage also presents a unique challenge. Traditional methods relying on local instance storage can fail when instances are terminated unexpectedly. This makes external backup and recovery systems essential to ensure data integrity and continuity[5].

Pricing Volatility

The pricing model for spot instances is another hurdle. Since prices fluctuate based on supply and demand, budgeting can become unpredictable. Unlike the fixed costs of on-demand or reserved instances, spot prices can vary significantly depending on the instance type, availability zone, and region. For instance, an instance that costs £0.05 per hour one week might spike to £0.30 per hour during high demand periods.

Adding to this complexity is the bidding system. Teams need to constantly monitor market trends and adjust their bids. A bid set too low increases the likelihood of interruptions, while a high bid reduces the cost advantage that makes spot instances appealing in the first place.

Engineering and Operational Complexity

Adopting spot instances also requires significant operational adjustments. Traditional monitoring tools designed for stable infrastructure often fall short in handling the ephemeral nature of spot instances. This demands the introduction of new alerting systems, health checks, and automated recovery processes.

Automation becomes indispensable, as manual intervention is impractical when instances can disappear without warning. This often involves deploying orchestration tools, automated failover systems, and self-healing infrastructure that can respond faster than human operators.

Integrating spot instances into existing CI/CD pipelines and workflows can also be challenging. Many legacy systems are built around stable infrastructure and may need substantial refactoring to handle the dynamic nature of spot instances. Diversifying instance types across multiple availability zones can add operational complexity but is often necessary to ensure consistent capacity. Testing and validation processes must also evolve to include simulations of interruptions and failover scenarios, often through chaos engineering techniques.

These challenges mean that adopting spot instances often takes more time and resources than initially expected. However, businesses that successfully address these obstacles can reap significant rewards. For example, CattleEye has managed to run all its batch processing on spot instances, achieving over 60% savings on EC2 costs[9]. By overcoming these challenges, organisations can unlock the potential of spot instances, as we will see in the next section.

6 Case Studies: Real-World Implementation Lessons

Real-world examples show how organisations have tackled the challenges of adopting spot instances while reaping significant cost benefits. These cases highlight practical strategies to manage risks like interruptions, pricing fluctuations, and operational hurdles, providing valuable lessons for businesses considering similar transitions.

Data Warehouse Cost Reduction

One organisation slashed its data warehouse compute costs by 70% through a focused spot instance strategy. They used spot fleets and spot blocks to ensure steady processing capacity while minimising interruptions. By diversifying across multiple instance types and availability zones - following capacity-optimised practices - they created fallback systems that seamlessly switched to on-demand instances when needed. Additionally, by integrating checkpointing into their ETL processes, they allowed jobs to resume from the last saved state instead of starting over after an interruption. This approach balanced cost savings with operational reliability.

Infrastructure Cost Savings in Continuous Monitoring

A content management provider significantly reduced infrastructure costs by pairing spot instances with continuous monitoring and targeted team training. They deployed real-time performance monitoring systems to track both costs and application metrics, enabling smarter decisions about instance selection and bidding. Training sessions equipped their engineering teams to manage interruptions and automate recovery processes. Regular evaluations of workload suitability ensured that only appropriate tasks were assigned to spot instances, while critical workloads remained on stable infrastructure. This methodical approach led to notable monthly savings and streamlined operations.

CoSpot Framework for Resource Allocation

A research initiative introduced the CoSpot framework, a method for cooperative resource allocation that balanced cost efficiency with performance needs. Using intelligent algorithms, the framework analysed demand patterns and historical usage data to predict the availability of spot instances. By adopting cooperative allocation strategies, they maintained both resource availability and service quality. This research-driven approach demonstrated how spot instance management can go beyond cost-cutting to become a strategic tool for resource planning.

NFL Season Scheduling

The National Football League (NFL) provides a striking example of spot instance adoption. By leveraging 4,000 EC2 Spot Instances across more than 20 instance types, they saved approximately £1.6 million per season [13]. Their system, designed to manage complex and time-sensitive workloads like season scheduling, effectively handled pricing volatility through well-planned resource allocation.

Freshworks Migration Strategy

Freshworks reduced infrastructure costs by 65% compared to on-demand instances after transitioning to managed spot instance solutions in 2016 [12]. Pradeep Thangavel, an engineering manager at Freshworks, shared:

In the beginning, we had never allocated a budget for our infrastructure costs, as it was considered part of the operational costs of running our applications, but as the company grew, we realised that cost efficiency is becoming a necessity when running at scale [12]

Freshworks faced challenges integrating spot instances with AWS OpsWorks. As Thangavel noted:

After thorough research, we came to realise that reliably managing spot instances is a massive automation challenge for us [12]

Their structured approach overcame these hurdles, improving interruption management and operational efficiency.

Amaysim Performance Optimisation

Amaysim, an Australian mobile virtual network operator, achieved a 75% reduction in compute costs by adopting architectural improvements and diversifying their instance types. They reduced batch processing times from 7 minutes to just 10–12 seconds [4]. Isaac Gittins, Cloud Architect at Amaysim, highlighted the benefits:

We've improved the resilience of our application and reduced the chance of outages using diversified Amazon EC2 Spot Instances [4]

This redesign not only cut costs but also enhanced system reliability by addressing interruption risks through thoughtful architecture.

These case studies make it clear that adopting spot instances isn’t as simple as switching instance types. Success often requires rethinking architecture, refining operations, and fostering a shift in how engineering teams approach infrastructure. Organisations willing to make these changes can unlock significant cost savings while improving the resilience of their systems.

Key Lessons and Best Practices

Drawing from challenges and real-world examples, these lessons offer a roadmap for effectively adopting Spot Instances while maximising cost efficiency, reliability, and performance.

Design for Interruption Tolerance

When working with Spot Instances, interruptions are inevitable. While fewer than 5% of Spot Instances are interrupted by EC2 before customers terminate them intentionally [10], successful organisations design their systems with these disruptions in mind.

One key strategy is implementing checkpointing, which allows workloads to resume from the last saved state [10]. Freshworks encountered this challenge firsthand. Pradeep Thangavel, Engineering Manager at Freshworks, highlighted:

The main challenge for us was to integrate both Spot Instances and AWS OpsWorks to work together because each has its own lifecycle. [12]

Building fault-tolerant architecture is critical. AWS's Well-Architected Framework principles provide a solid foundation for this [10]. Delivery Hero serves as an example, having designed their systems to handle 4×–5× traffic spikes while running 90% of their Kubernetes workloads on Spot Instances. This was achieved by focusing on application resilience and using multiple instance redundancies [5].

AWS also provides a two-minute interruption notice, which can be leveraged for controlled shutdowns [10]. For instance, Kubernetes users can utilise PreStop hooks to execute orderly shutdown procedures and preserve state during this window [11]. Additionally, querying the Instance Metadata Service allows teams to detect interruptions and trigger automated responses via Amazon CloudWatch Events or Amazon EventBridge [10].

These practices form the basis for more automated and diversified strategies.

Use Automation and Diversification

Automation plays a vital role in managing Spot Instances, especially given their potential interruptions and fluctuating pricing [15]. Combined with resource diversification, automation ensures stability and optimised costs.

Diversifying resources across multiple instance types and Availability Zones reduces risk [6]. A capacity-optimised allocation strategy can help teams launch instances from Spot pools with the most available capacity [10]. By selecting instance types from various families, sizes, and zones, organisations maintain flexibility and minimise disruptions [10].

Using attribute-based selection in EC2 Auto Scaling or Fleet simplifies the process of matching instances to required vCPUs, memory, and storage [6]. This eliminates the need for manual instance selection while ensuring performance standards are met.

Another effective approach is the price-capacity-optimised allocation strategy within Auto Scaling groups and EC2 Fleet. This provisions instances from the most available Spot pools at the lowest cost [6]. For example, Yotpo automates the entire Spot Instance lifecycle, running at least 80% of its workloads on Spot Instances [15].

Spot placement scores are another helpful tool, guiding teams to Regions and Zones with optimal capacity [6]. When combined with a diversified strategy, this approach spreads Spot Instances across multiple pools, improving availability and reducing interruptions [14].

Monitor and Optimise Continuously

After implementing robust design and automation strategies, continuous monitoring ensures long-term success. Keeping an eye on pricing, availability, and interruption rates is essential for efficient Spot Instance management [5].

Porter’s experience with Ocean highlights the importance of monitoring tools. As Jijo T. Joy, Senior DevOps Engineer at Porter, explained:

Ocean's logging and recommendations were valuable in understanding infrastructure consumption and costs, providing ECS service-level cost visibility not directly available in AWS. [16]

Adjusting strategies to reflect market conditions is equally critical. ITV’s phased migration over 18 months gradually increased Spot usage from 9% to 24%, with constant monitoring of performance and costs along the way [5].

Metrics like load balancer connections and auto-scaling group performance offer insights into how well interruption-handling mechanisms are working [10]. These metrics help identify when adjustments are necessary to maintain fault tolerance.

Ultimately, the most successful organisations treat Spot Instance management as an ongoing effort. Regularly reviewing usage patterns, refining bidding strategies, and updating automation ensures systems remain aligned with business needs and market trends.

Spot Instances vs. Other Instance Types

Choosing the right EC2 instance type is a balancing act between cost, reliability, and operational complexity. Each option caters to specific needs, and the decision often comes down to weighing cost savings against the demands of maintaining reliability and managing workloads.

Cost considerations play a major role. Spot Instances stand out for their affordability, offering savings of 70–90% compared to On-Demand instances [2]. Reserved Instances also provide substantial discounts - up to 72% lower than On-Demand pricing [2][17] - but require a longer-term commitment. On-Demand instances, while the most flexible with their pay-as-you-go model, come at a higher hourly rate [18]. These cost differences often shape the strategies companies adopt.

For example, a mid-sized financial services firm implemented a hybrid approach to optimise costs. They allocated 60% of their workload to Reserved Instances, achieving a 40% saving; 25% to Spot Instances, saving 85%; and kept 15% on On-Demand for critical, unpredictable tasks. This strategy led to a 62% reduction in annual EC2 costs, amounting to savings of approximately £155,000 [18].

Reliability and the risk of interruptions are also key factors. Spot Instances, while economical, come with the caveat of potential interruptions, with a two-minute termination warning. However, less than 5% of Spot Instances are interrupted in a typical month [19]. In contrast, both On-Demand and Reserved Instances offer uninterrupted compute power, making them more reliable for critical workloads [1].

Engineering complexity varies significantly among the instance types. Spot Instances require additional engineering effort to handle interruptions, diversify capacity, and implement autoscaling [2]. Reserved Instances are simpler to manage due to their predictable performance, though there’s a risk of underutilisation [2]. On-Demand Instances involve the least engineering effort but come at the highest cost [3].

Comparison Table

This breakdown highlights the importance of tailoring your strategy to your workload needs. Spot Instances excel for flexible, fault-tolerant workloads, while Reserved Instances are ideal for predictable, steady-state operations. On-Demand Instances remain the go-to for unpredictable or short-term requirements. By combining these options in a hybrid strategy, organisations can optimise costs while maintaining reliability and operational efficiency [2].

Conclusion: Maximising Cloud Cost Savings

The case studies discussed reveal that achieving success with spot instances relies on three key elements: resilience, automation, and continuous monitoring. For example, one organisation reduced costs by an impressive 75% while simultaneously boosting performance and availability, all by adopting these principles [4]. Their approach included designing fault-tolerant applications and diversifying instance types across multiple availability zones.

Resilience forms the backbone of any effective spot instance strategy. Businesses must prepare for potential interruptions by implementing techniques like checkpointing to save progress externally, leveraging spot-integrated services such as Amazon ECS and AWS Batch, and adopting capacity-optimised allocation strategies [10]. These measures ensure that workloads can handle disruptions without compromising functionality.

Automation is another critical factor. It simplifies the often-complex task of managing spot instances. For instance, Freshworks achieved average savings of 65% by deploying automated systems that managed the entire lifecycle of spot instances - from selecting cost-effective options to transitioning workloads seamlessly when capacity became unavailable [12][15].

Equally important is continuous monitoring. This involves tracking costs, performance metrics, interruption rates, and recovery times to refine strategies over time [20]. By analysing spot price history, setting up alerts for price changes, and using tools like AWS Spot Instance Advisor, businesses can make smarter, data-driven decisions [7].

The potential savings are substantial - up to 90% compared to on-demand pricing [6][8]. However, realising these savings requires expertise in cloud architecture, automation, and ongoing optimisation. For organisations aiming to implement these strategies effectively, working with specialists in cloud cost management can make a significant difference. For example, Hokstad Consulting offers tailored solutions, helping businesses cut costs by 30–50% through strategic spot instance use, automated management, and detailed monitoring.

To maximise cloud cost savings, the formula is straightforward: design resilient systems, automate management processes, and monitor continuously. Companies that embrace these practices can achieve substantial cost reductions while maintaining the reliability and performance their operations demand.

FAQs