ABP Live served 2.08 million concurrent requests at 100% uptime on AWS

ABP Live expected campaign traffic to surge to two million concurrent requests within short windows. Conventional scaling sized for that peak would sit idle and expensive for the rest of the year, and the platform still had to scale by itself for the next event.

• Scalability: The existing infrastructure could not handle the anticipated surge of 2 million concurrent requests.
• Cost efficiency: Traditional scaling was too expensive for infrequent, short traffic spikes.
• Agility: The platform needed to scale automatically for future high-demand events.

BluePi built a serverless request path on AWS. Amazon API Gateway receives every call, AWS Lambda runs the application code with no servers to manage, and Amazon DynamoDB stores campaign data at whatever volume the traffic demands. Amazon CloudWatch tracks performance and resource use, and Amazon SNS carries real-time notifications for event triggers and scaling actions.

• AWS API Gateway: A single entry point manages all incoming API requests.
• AWS Lambda: Runs the application code with no servers to manage.
• AWS DynamoDB: Stores campaign data in a NoSQL database that scales with traffic.
• AWS CloudWatch: Monitors application performance and resource use.
• AWS SNS: Sends real-time notifications for event triggers and scaling actions.

The architecture served 2.08 million concurrent requests at 100 percent uptime during high-demand events. Campaign infrastructure cost fell from $530 to $15, and the team ships features without server management work.

• Peak load handled: The serverless architecture handled 2.08 million concurrent requests with 100% uptime.
• Cost savings: Campaign cost fell from $530 to $15.
• Improved agility: The team develops and deploys new features without server management overhead.
• Reduced maintenance: No servers to provision, manage, or maintain.
• Lower total cost of ownership (TCO): Costs for server management, software licenses, and infrastructure maintenance went away.

The traffic pattern ruled out provisioned capacity: rare, short spikes separated by long quiet periods. Every component had to scale on demand and bill only for actual use. The path also had to stay simple enough for the team to operate without a dedicated infrastructure function.

The platform was judged on the conditions that matter for a live campaign: concurrent-request capacity, availability through the surge, and cost per campaign. CloudWatch metrics gave the team a continuous view of performance and resource consumption during each event.

Capacity planning stopped being a pre-campaign project. The team develops and deploys features without provisioning work, and per-campaign infrastructure spending dropped by two orders of magnitude, which changed which campaigns were worth running.