Abstract

As a reaction to the recent “Big Data” trend, a new breed of systems for scalable data processing has emerged. Our system, Stratosphere, offers an extensible query language for posing queries on complex nested data, an efficient processing engine designed to scale on very large clusters and leverage cloud elasticity, as well as a query optimizer and a runtime engine that guarantee the efficient execution of queries, including iterative queries. Stratosphere pushes the MapReduce paradigm forward by incorporating several optimizations known from parallel databases, as well as novel techniques, while retaining the flexibility of in-situ processing of data using complex user-defined functions.

In this talk, I will provide an overview of the Stratosphere system, placing emphasis on how to optimize and execute in parallel an extended dataflow programming model with user-defined functions and iterative constructs. I will then provide a research outlook for scalable data analytics that includes research topics in the intersection of programming languages, databases, and networks.