赞助商：AWS-了解如何使用 AWS 上的 Databricks 管理数据和 AI 工作负载的成本.pdf

1、2024 Databricks Inc.All rights reservedVenkat Viswanathan(AWS)and Jeroen Meulemans(Databricks)June 12th20241Learn how to manage Learn how to manage cost for data and AI cost for data and AI workloads with workloads with Databricks on AWSDatabricks on AWS2024 Databricks Inc.All rights reservedAgendaC

2、ustomer challengesPrinciples of cost optimizationBest practicesAdopt AWS cloud financial management(CFM)22024 Databricks Inc.All rights reservedTrack and interpret usage across regions,workspaces and accountsTake corrective action through identifying anti patterns and outliers Global usage trackingM

3、easuring valueGovernanceTrack return on investments-ensuring that projects deliver more value than they spend to runObjectively justify spend for early stage developments and deploymentsCoEs want to maximize their efforts and articulate the value they bring to teamsEstablish monitoring and enforcing

4、 behaviors across diverse requirements3Customer challenges2024 Databricks Inc.All rights reservedPrinciples of cost optimization1.Choose optimal resources.2.Dynamically allocate resources.3.Monitor and control cost.42024 Databricks Inc.All rights reservedSimplify your right sizing journey with AWS C

5、ompute OptimizerMake the right choice12Saves time comparing and selecting optimal resources for your workloadContinuously scan your resource usage and match your workload to optimal resourcesApplies insights from millions of workloads to make recommendations Ensure you are consistently making the ri

6、ght choice52024 Databricks Inc.All rights reservedExample:Amazon EC2 instancesDatabricks workload 40%CPU during the day 10%CPU during the night 30%RAM usage 1 Mbps network usage more than 99 percent of the time=current_timestamp()-INTERVAL 1 day;ZORDER BY(eventType);Hive-style partitioning is a hier