HadoopDB therefore pushes computation closer to data (into the data tier) to achieve maximum parallelization in a multi-node clustercomplexity of the data tier and its parallel nature is hidden from the application developer
Universal Protein Resource.presentation layer consists of a web-based interface where analysts specify queries and view resultslogic layer consists of a SPARQL to SQL conversion toollogic and data layer communicate through JDBC
presentations provide our audience with an idea of the eort required for data preparation in HadoopDB
HadoopDB in Action
Introduction Architecture and Design Example application Demostration Scenario
Managing and analysing massive data ◦ Provides high performance ◦ Scales over clusters of thousands of heterogeneous machines ◦ Versatile-adaptability of a system to analytical queries of varying complexityHow does one build real world applications withHadoopDB?
Database Connector - connects Hadoop with the single-node database systems. Data Loader - partitions data and manages parallel loading of data into the database systems. Catalog - tracks locations of different data chunks,including those replicated across multiple nodes. SQL-MapReduce-SQL (SMS) planner - extends Hive to provide a SQL interface to HadoopDB
Supports any JDBC-compliant database server as an underlying DBMS layer Applications built on top of HadoopDB generally use the 3-tier architecture ◦ data tier ◦ business logic tier ◦ presentation tier HadoopDB is a black box(in application perspective)
A semantic web/biological data analysis application. A business data warehousing application.
Semantic web is an effort by the W3C to enable integration and sharing of data across dierent applications RDF- is a directed, labeled graph data format for representing information in the Web SPARQL –is an RDF query language
Find all proteins whose existence in the `Human organism is uncertain SPARQL query :
demonstrate ◦ how the data administrator should prepare the dataset. Analyst- is shielded from the complexity of the actual implementation of the RDF storage layer.
Natural target application for HadoopDB. Common business data warehousing workloads are read-mostly and involve analytical queries over a complex schema To achieve good query performance, the dataset requires signicant preparation through data partitioning and replication to optimize for join queries Data & Queries- TPC-H benchmark
Audience is invited to query both data sets through HadoopDB Data sets are located in a remote cluster Multiple users interaction- two client machines that connect to the clusters.
user selects dataset SemanticWeb—Biological Data Analysis - An animation of the behind-the-scenes data preparation & loading is presented - Details on the tools used for data conversion from RDF to relational form. Business Data Warehousing- the animation provides details on the partitioning scheme, the interaction between the loader and catalog components, and a summary of the configuration parameters User select and parametrize a query to execute -User can then monitor the progress of query execution
In addition demonstrate HadoopDBs fault- tolerance with the introduction of a node failure. For a subset of the predened queries, as the query executes in the background, an animation of the flow of data and control through the HadoopDB system is simultaneously presented, highlighting which parts of the query execution are run in parallel.