Databricks SQL 上的快速查询、高并发和扩展的性能最佳实践.pdf

《Databricks SQL 上的快速查询、高并发和扩展的性能最佳实践.pdf》由会员分享，可在线阅读，更多相关《Databricks SQL 上的快速查询、高并发和扩展的性能最佳实践.pdf（34页珍藏版）》请在三个皮匠报告上搜索。

1、DBSQL Serverless Best Practices for the best Performance Data&AI Summit 2025Jeremy Lewallen&Mostafa MokhtarHow far weve come!Performance Feature AnnouncementsBest Practices for Best PerformanceQuery Troubleshooting for Best PerformanceAgendaWe havent always been able to ask any question of our data2

2、010:The Dark Ages of Data.When Most Questions Couldnt Be AnsweredNo Spark.No Lakehouse.Just Silos and Struggle.Most Questions Couldnt Be AnsweredData Engineering Was PrimitiveThen,Spark Changed Everything2020:A Decade Later,the economic wall emergesThe data got bigger,the questions got harder,and th

3、e cost explodedThe Costs ExplodedThe Questions Got Harder+=The Data Grew ExponentiallyThe traditional model is,Throw more hardware The traditional model is,Throw more hardware at the problemat the problemTraditional warehouses have hard fixed boundaries;Costs expand at least linearlyBut,this isnt 20

4、20,the modern warehouse must do many things wellThe Architecture of Intelligence:This is how we think about performance in DBSQLOne goal:Sub-linear scaling The Economic BreakthroughLayer 1:Data IntelligenceLearns patterns and reorganizes data automatically for optimal accessLayer 2:Execution Intelli

5、genceRewrite query plans on-the-fly using Optimizer,and execute rapidly with the Photon engineLayer 3:Operational IntelligencePredicts workloads,adds/removes compute,and self-optimizes continuouslyMultiple Layers of IntelligenceDBSQL Serverless is 2.5x better TCO AND 3x faster across all scale facto

6、rs 1TB+As data&usage grow,the economics become better for the customerOne Warehouse Engine,All WorkloadsDashboards,pipelines,apps,and AI all run on the same lakehouse engineno silos,no trade-offs.Scale to your entire companyServerless and rapid scaling hold latencies steady even as user demand spike