Optimizing Price Performance in AdTech Using Amazon EC2 Spot Instances with Epsilon

Founded in 1969, Epsilon connects the world’s leading advertisers with consumers to drive performance while respecting and protecting consumer privacy and client data. These capabilities are powered by on-premises and AWS cloud infrastructure. Every day, the company handles a massive volume of data transactions and requests, which can be as high as 100 TB per client.

“On the AdTech side, we see more than 700 billion requests per day for bid stream processing,” says Chris Wissing, senior vice president of technology at Epsilon. “We typically work with our clients’ first-party data or a subset of data that we have created and procured over time. This amounts to tens of millions of transactions per day for a typical large client.”

The scale of Epsilon’s operations not only required significant compute resources but also made it challenging to optimize cost and performance. In 2010, the company began to migrate away from its exclusively on-premises infrastructure to the cloud, and adopted various Amazon EC2 instances to manage its growing workload demands.

Although Amazon EC2 instances offered scalability and flexibility, Epsilon’s data volume continued to expand over time. The scope of its operations changed due to frequent mergers and the introduction of new business objectives, which required a reassessment of its workloads to maintain optimal efficiency. Epsilon also needed to manage resources across different on-premises and cloud environments, which introduced complexities and sometimes led to inefficiencies. To optimize for price performance, it began to reevaluate its use of Amazon EC2 resources.

To identify areas for optimization, Epsilon conducted an in-depth assessment of its existing workloads. It first categorized the workloads that could tolerate potential interruptions without affecting overall service and moved them to Amazon EC2 Spot Instances, which are available at up to 90 percent discount compared to On-Demand pricing. Because Amazon EC2 Spot Instances have a 2-minute warning before interruptions, the company determined that they were an optimal choice for fault-tolerant and stateless workloads. The strategic use of Amazon EC2 Spot Instances helped Epsilon achieve significant cost savings while maintaining operational flexibility.

To further optimize and diversify its Amazon EC2 choices, Epsilon introduced AWS Graviton-based instances to its compute fleet. The company first targeted AWS managed services such as Amazon ElastiCache—a serverless, Redis- and Memcached-compatible caching service—because they were the simplest to transition. AWS handles software compatibility, security, and other operational overhead for these services, simplifying the migration process. Then, the company migrated compute-intensive workloads that required more consistent performance and availability. Epsilon was able to migrate one of its largest data processing jobs to AWS Graviton-based Amazon EC2 Spot Instances, reducing the scan time by nearly 20 percent.

“Since we started this journey, most of our data engineering processing, which we run on Spark, is now on AWS,” says Gairik Chakraborty, vice president of database technology at Epsilon. “From an analytics, data engineering, and data processing perspective, around 50–60 percent of our workloads have already been moved to AWS Graviton processors.”

Throughout this journey, Epsilon continually optimized its AWS usage. The company monitored its deployments and adjusted its strategies to optimize for cost and performance, which led to ongoing improvements. These optimizations resulted in an average 15 percent cost reduction using AWS Graviton-based instances, as well as improved performance. “We have seen a 15–20 percent reduction in runtime and a 5–10 percent reduction in CPU usage using AWS Graviton-based instances,” says Chakraborty.

With these improvements, Epsilon has been able to better manage its costs and lower its total cost of ownership. These compute cost savings have facilitated the company’s expansion into the midmarket sector, fueling its continued growth. “On AWS, we can invest more in our products and less on the compute itself,” says Wissing. “We’ve been able to focus more on projects that ultimately benefit our clients, such as application and product development.”

On AWS, Epsilon has achieved significant cost and performance efficiencies, which translates into enhanced services for its clients. Thanks to the success of this project, the company continues to use AWS Graviton-based instances for suitable workloads within their organization. For example, the majority of Epsilon’s functions on AWS Lambda, a serverless, event-driven compute service, are powered by AWS Graviton processors.

Looking ahead, Epsilon plans to explore AWS Graviton4-based instances for greater efficiency and performance gains. It is also exploring ways to move workloads running on Amazon Relational Database Service (Amazon RDS)—an easy-to-manage relational database service—and Amazon Elastic Kubernetes Service (Amazon EKS)—the most trusted way to start, run, and scale Kubernetes—to AWS Graviton processors.

“Our ultimate goal is to build the best software and products for our clients,” says Wissing. “The cloud will have a huge impact on how we optimize our business and reach our goal in the future.”