Unified analytics platform for Bigdata
Submitted by Amareshwari Sriramadasu (@amareshwari) on Monday, 12 May 2014
Section: Full talk Technical level: Intermediate
This talk is about a system developed at InMobi to support OLAP data cubes on top of Hive metastore. With this abstraction, users can reference single schema and data stored across diverse storage engine and that users can query data on the logical tables without knowing about schema details like relationships, rollup levels, data location and data types.
Conventional columnar databases (RDBMS) systems lend themselves well for interactive SQL queries over reasonably small datasets in the order of 10-100s of GB, while hadoop based warehouses operate well over large datasets in the order of TBs and PBs and scales fairly linearly. Though there have been some improvements recently in storage structures in the Hadoop warehouses such as ORC, queries over hadoop still typically adopts a full scan approach. Choosing between these different data stores based on cost of storage, concurrency, scalability and performance is fairly complex and not easy for most users. This talk presents Grill, the new analytics platform for InMobi, a system built at InMobi to precisely solve this problem on top Hive metastore.
The Hive metastore in its current state allows users to represent structured data in simple tables. However, it does not allow expressing relationships or richer DWH concepts like facts, dimensions and etc. With Hive data cubes, users can query data stored in HDFS, S3, Redshift etc, with a single query language and schema. Underlying execution engines like Hive, Impala, Shark etc can be plugged in and utilized at run time. The execution engine used is transparent to the user. The system provides a unified logical schema to users consisting of cubes, facts and dimensions; and users can issue queries at a conceptual level without knowing about roll-up intervals, partitions, data types, underlying storage and table relationships; they will be figured out automatically.
Amareshwari is currently working as Architect in platform team at Inmobi, where she works on Hadoop and related projects for data collection and analytics. She is member of Apache Hadoop PMC and is Apache Hive committer. She has been working on Hadoop and its eco system since 2007. Prior to Inmobi, she was working with Yahoo! in core Hadoop team. She holds bachelor's degree in computer science and engineering from National institute of technology, Waragal, India; and master's degree in Internet science and engineering from Indian Institute of Science (IISc), Bangalore, India.