About This Architecture

BMW's real-time data platform separates hot and cold processing paths to handle 200 records/second from plant floor sources including PI Historian, Ampla IQT, DAS, and Citect SCADA. The hot path (<5s) streams events through Kinesis to Lambda consumers routing to DynamoDB (KPI dashboards), Timestream (trend charts), and EventBridge (alerts), while the cold path batches to S3 via Firehose for audit and historical analysis. AWS Glue 5.1 with Step Functions orchestrates multi-zone ETL (raw, curated, refined) feeding Athena, QuickSight, and SageMaker for analytics and ML, with Lake Formation governance and AppSync WebSocket push for real-time frontend updates. This dual-path architecture demonstrates how to balance sub-second operational dashboards with cost-effective historical analytics and ML feature engineering at industrial scale. Fork this diagram on Diagrams.so to customize for your multi-site deployment, adjust shard counts and batch windows, or integrate IoT SiteWise for Phase 1B expansion.

People also ask

How do you design an AWS data platform that serves both real-time operational dashboards and historical analytics without sacrificing cost or latency?

This diagram shows a dual hot/cold path architecture: the hot path streams 200 rec/s through Kinesis to Lambda consumers routing to DynamoDB (<10ms KPI reads) and Timestream (<2s trend queries), while the cold path batches to S3 via Firehose for cost-effective audit and ML. AWS Glue 5.1 orchestrates multi-zone ETL (raw→curated→refined) feeding Athena, QuickSight, and SageMaker, with Lake Formation