Collaborative Research: PPoSS: A Full-stack Approach to Declarative Analytics at Scale

点击翻译按钮获取中文摘要

Declarative programming languages permit users to define a problem's rules and goals, and the structure of a valid solution, automating the mechanics of computation within their implementations. Such expressive programming systems provide an opportunity for non-experts to immediately scale their analytic tasks to massive datasets, leveraging AI-based programming. The project's novelties are a set of techniques, integrating novel optimizations across the full computation stack, that deliver orders-of-magnitude scalability enhancements to declarative programming. The project's impacts are centered on permitting non-specialists to scale sophisticated deductive inference algorithms to the next generation of cloud-based clusters and supercomputers.Linguistically, the project's approach is based on a key semantic extension to Datalog to support indexing for structured inductive data. While algebraic data is supported in currently existing Datalog engines (e.g., Souffle), the project’s novel approach also materializes indices for all such ADTs, enabling orders-of-magnitude algorithmic improvements in runtimes of queries over algebraic data. Operationally, the project advances state-of-the-art implementation strategies based on parallel relational algebra, which enables off-the-shelf data parallelism that rapidly scales to many-core clusters and supercomputers via MPI. The project seeks to integrate each of these technologies to scale key applications---including program analysis and security auditing---and demonstrate their application to large datasets enabled via the project’s unique synthesis of these technologies.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.