Towards Complex Query Processing over Key-Value Cloud Stores

Опубликовано 12 августа 2016, 0:27

Facts: Cloud infrastructures bear an ever-increasing responsibility for storing and maintaining massive volumes of data for different types of data-intensive applications. Key-value cloud-stores, have become a premium choice as the storage back-end for such applications. We need complex query processing capability to access/analyze this data. Questions: Do we have adequate solutions required to support complex queries, over data residing in such storage infrastructures? Do standard, ΓÇ£cloud-friendlyΓÇ¥ approaches, such as MapReduce-based algorithms, offer a satisfactory solution? What additional support, in the form of indexing and query processing algorithms, would expedite query processing? Can we do so, while benefiting from the simplicity of the key-value systems' interface and free-ride on their inherent scalability, elasticity, and reliability? Answers: In this talk I will present novel indexing structures and processing algorithms for complex query types. Specifically, I will first cover interval queries in depth, presenting indices and associated query processing algorithms. I will also overview indexing and query processing approaches for rank-join queries. Our contributions include key-value representations of our index and statistical structures, MapReduce algorithms to build and populate them, and query processing algorithms utilizing them, catering to idiosyncrasies of key-value stores, but inheriting their advantages. Our implementation and experimentation are over the popular HBase key-value store. I will report on the results of extensive performance evaluations, which show large performance improvements. En route, I will touch upon differences in existing key-value system architectures and their implications. The talk will conclude with the lessons we have learned, pointing to key design decisions, and promising ideas for outstanding challenges.