AWS re:Invent 2025 - Accelerating the Connected Future: EDA for the unpredictable (IND308)

Accelerating the Connected Future: BMW's Event-Driven Architecture for Unpredictable Workloads

Overview

• BMW is transforming its vehicle fleet into intelligent, connected companions with its BMW Connected Drive platform
• The remote services feature of Connected Drive allows customers to remotely control their vehicles from anywhere
• BMW's remote services back-end was facing challenges with scalability, reliability, time-to-market, and cost as the connected vehicle fleet and data demands grew

Legacy Architecture and Challenges

• In 2023, the remote services back-end was running on a monolithic architecture in AWS, using Amazon Kubernetes Service and Amazon RDS
• While the system was functional, it faced issues with:
  • Increasing technical debt and complexity over time
  • Difficulty scaling to handle tripling request volumes in 2 years
  • Slow time-to-market for new features and vehicle models
  • Concerns about maintaining reliability and low latency under high load
  • Lack of cost optimization for the growing AWS infrastructure

Redesigning Remote Services with Serverless

• BMW partnered with AWS to reimagine the remote services architecture using a serverless, event-driven approach
• Key goals were to increase time-to-market, reduce maintenance effort, improve scalability and reliability, and optimize costs
• The new architecture broke down the monolith into independent microservices:
  • Event Creation: Receives requests from the My BMW app and validates/enriches them
  • Outgoing Vehicle Communication: Translates events into MQTT messages for the vehicle
  • Incoming Vehicle Communication: Processes asynchronous responses from the vehicle
  • Event Processing: Handles event distribution, persistence, and lifecycle management

Technical Details

• Used AWS Step Functions, Lambda, SNS, SQS, and DynamoDB to implement the event-driven architecture
• Leveraged AWS Fargate for the vehicle communication services to maintain bidirectional MQTT connectivity
• Implemented custom metrics using the Embedded Metric Format to track end-to-end latency
• Utilized subscription filters on SNS topics to avoid unnecessary processing in downstream components
• Employed a delayed queue mechanism to handle potential connectivity issues and ensure event completion

Optimizations and Iterations

1. MQTT Broker Integration: Moved from a direct Fargate connection to the MQTT broker to an abstracted API-based integration, reducing maintenance overhead.
2. Vehicle Simulator: Replaced the Fargate-based vehicle simulator with a serverless, configuration-driven solution, enabling extensive testing and validation.
3. Blue-Green Deployments: Implemented an Application Load Balancer-based approach to enable safe, canary-style deployments without disrupting customer traffic.
4. Developer Decoupling: Provided each developer with their own set of duplicated components, allowing for independent testing and deployment without conflicts.

Business Impact and Results

• Increased time-to-market for new features and vehicle models by 60%
• Significantly reduced maintenance effort and costs by leveraging fully managed AWS services
• Achieved almost limitless scalability to handle the growing connected vehicle fleet and data demands
• Maintained low latency (P99 < 1 second, P50 < 400ms) even under sustained high load (100 requests/second for 3 hours)
• Reduced overall AWS infrastructure costs by 20% through optimized serverless architecture

Conclusion

• BMW's transition to a serverless, event-driven architecture for its remote services back-end has enabled the company to address key challenges and unlock significant benefits:
  • Improved time-to-market and development velocity
  • Reduced maintenance overhead and costs
  • Enhanced scalability, reliability, and performance
  • Optimized AWS infrastructure costs
• The iterative approach and focus on developer experience further strengthened the solution, making it adaptable to future requirements and vehicle generations.