Efficient Column Store for Cloud Data Lakes

This lecture discusses the design of Pixels, a column store optimized for cloud data lakes like S3. It covers the challenges of using S3-like storage, the trade-offs between price and performance, and the strategies to mitigate latency and improve query performance. The lecture also presents the effects of various optimizations in Pixels, such as request merging, asynchronous reading, and straggler mitigation, through real-world workload evaluations and comparisons with Parquet. It concludes by highlighting the benefits of using a column store on S3-like storage, emphasizing the importance of reducing request latency and improving overall query efficiency.