BigObject Store: In-Place Computing for Interactive Analytics

摘要

In order to address real-time analytic problems for big data, we have introduced a technology, In-Place Computing, a set of principles for storing and computing big data. It defines an abstract model in which data objects live and work on a flat and infinite address space. The term in-place indicates data is ready for computing in two ways: (1) in-memory and data-centric, in which computations take place where data resides, and (2) well-organized data objects, which have algebraic expressions and operators. Based on these principles, we have developed an in-place computing system, BigObject Store, which implements an extended relational model in the sense allowing various shapes of macro objects besides tables and meanwhile supporting algebraic operators besides join. An abstraction, Macro Data Structure, is defined and a programming paradigm is used to support data-centric computing. Our experiments show that BigObject delivers promising results on typical analytic tasks.