CRUCIBLE: towards unified secure on- and off-line analytics at scale

Abstract

The burgeoning field of data science benefits from the application of a variety of analytic models and techniques to the oft-cited problems of large volume, high velocity data rates, and significant variety in data structure and semantics. Many approaches make use of common analytic techniques in either a streaming or batch processing paradigm. This paper presents progress in developing a framework for the analysis of large-scale datasets using both of these pools of techniques in a unified manner. This includes: (1) a Domain Specific Language (DSL) for describing analyses as a set of Communicating Sequential Processes, fully integrated with the Java type system, including an Integrated Development Environment (IDE) and a compiler which builds idiomatic Java; (2) a runtime model for execution of an analytic in both streaming and batch environments; and (3) a novel approach to automated management of cell-level security labels, applied uniformly across all runtimes. The paper concludes with a demonstration of the successful use of this system with a sample workload developed in (1), and an analysis of the performance characteristics of each of the runtimes described in (2).