Laputa is a security framework for Apache Spark that enforces fine-grained data access policies and protects against malicious users and cloud managers. It uses pattern matching to check policies at the physical plan level and leverages confidential computing to compartmentalize Spark applications, ensuring the integrity of the entire data analysis pipeline. Evaluation on TPC-H benchmarks and real-world ML applications shows Laputa successfully blocks malicious applications with moderate performance overhead while requiring minimal modifications to existing Spark code.
196 Impressions