Reliable scalability: How Amazon.com scales on AWS

How does Amazon handle 126 million requests per second on Prime Day? Discover the cell-based architecture and serverless patterns that make it possible.

#1about 4 minutes

From a single monolith to a service-oriented architecture

Amazon's initial architecture consisted of a single C/C++ binary and a large database, which was broken down into a service-oriented architecture to enable faster development and scaling.

#2about 3 minutes

Achieving massive scale on AWS for Prime Day

Prime Day traffic, with DynamoDB handling 126 million requests per second, illustrates the need for reliability, which is achieved through the principles of the AWS Well-Architected Framework.

#3about 3 minutes

How IMDb uses serverless microservices and GraphQL

IMDb migrated from a monolith to AWS Lambda, segmenting its workload into distinct microservices that are unified into a single API endpoint using GraphQL and a federated schema.

#4about 2 minutes

IMDb's gateway architecture and two-pizza team ownership

IMDb's backend uses a gateway-based architecture with a schema manager, where each microservice is fully owned and operated by a small, autonomous 'two-pizza team'.

#5about 3 minutes

Scaling AWS Lambda to handle high traffic spikes

To handle 800,000 requests per minute, IMDb uses AWS Lambda's auto-scaling capabilities and provisioned concurrency to keep functions warm and eliminate cold start latency.

#6about 2 minutes

Using AWS WAF and CloudFront for security and performance

IMDb improves reliability and performance by using AWS Web Application Firewall (WAF) to block malicious bot traffic and Amazon CloudFront CDN to reduce latency for global users.

#7about 5 minutes

Implementing fault isolation with a cell-based architecture

Amazon's fulfillment centers use a cell-based architecture to achieve fault isolation, ensuring that a failure in one cell does not impact the operations of others.

#8about 1 minute

Managing cells with separate AWS accounts and assignment tools

Cells are implemented as separate AWS accounts to create hard boundaries, and a custom tool assigns fulfillment centers to cells to maintain balanced resource distribution.