Title

AWS re:Invent 2022 - Running high-throughput, real-time ad platforms in the cloud (ADM302)

Summary

• The session focused on the experience of building high-throughput, real-time ad platforms capable of processing over 1 trillion transactions per day.
• Victor Gershkovich from AppsFlyer and Akhil Andapalli from AWS shared insights on transitioning from EC2 to EKS, optimizing with Graviton instances, and empowering developers with self-serve infrastructure.
• AppsFlyer's journey involved moving from a classic EC2 approach to an EKS infrastructure, optimizing with Graviton instances for cost and performance, and providing developers with tools for easy infrastructure management.
• The presenters discussed the importance of stateless architecture, using spot instances, and leveraging Graviton processors for better cost efficiency.
• They shared best practices for reducing costs and improving performance, including minimizing boot time, using attribute-based instance selection, and monitoring spot placement scores.
• The session also covered the benefits of using Graviton processors, including significant cost savings and reduced energy consumption.
• Tips for optimizing code and infrastructure were provided, such as using fast JSON libraries, optimizing web server scalability, and employing non-blocking tasks to reduce latency.
• The session concluded with the announcement of an open-source real-time bidding application and an invitation for feedback.

Insights

• Transitioning to a stateless architecture and separating processors from data producers can significantly improve scalability and reduce costs.
• Utilizing spot instances and Graviton processors can lead to substantial cost savings, with Graviton offering up to 40% better price performance over similar x86 instances.
• Optimizing code by replacing standard libraries with more efficient alternatives can drastically reduce latency and increase throughput.
• Implementing non-blocking tasks and optimizing data streams can further reduce latency and improve overall system performance.
• The use of open-source tools and community contributions, such as the TerraCrust API for Terraform, can simplify infrastructure management for developers.
• The session highlighted the importance of continuous optimization and experimentation to achieve the best performance and cost efficiency in cloud-based ad platforms.
• The announcement of an open-source real-time bidding application indicates AWS's commitment to sharing knowledge and tools with the wider community, encouraging collaboration and innovation.