The Washington Semantic Web Meetup

If we are ever going to have a truly scalable Semantic Web (and, in particular, reasoning on a global scale), we will need to improve the performance of very large RDF sets. Now that we are starting to see trillion-triple data sets, it is time to begin applying massively parallel computing architectures to triple stores and reasoning systems. Hadoop is one possible framework that we can use, and some SemWeb projects have already been investigating it. 

This Hadoop meetup will be a combined meetup with the Hadoop User Group. We have two very special guests presenting, Tom White and Aaron Cordova. Tom has been instrumental in shaping the direction and development of Hadoop and many projects in Hadoop's ecosystem. He is also the author of the book "Hadoop: The Definitive Guide" and a part of the team at Cloudera. Aaron Cordova is a UMD graduate working at Booz Allen Hamilton. Before joining Booz Allen Aaron played a key role in defining the critical large scale data analytics infrastructure and applications at the NSA. Currently Aaron focuses on helping government organizations manage and analyze large amounts of data using technologies such as Hadoop, Hive, and HBase.

o  6:30 - 7:00 Food and Refreshments, Socialize
o  7:00 - 7:30 Tom White - Hadoop's powerful parallel processing paradigm provides a great generalized framework for storing and analyzing data. For all its raw power however the practical use of Hadoop requires more. What's needed is an integrated stack of components which makes it easier to develop and use real-world applications in a production environment. Tom will discuss the evolving Hadoop platform, its components and how each fills a critical role in making Hadoop more useful in the enterprise.
o 7:30 - 7:40 Short Break
o 7:40 - 8:10 Aaron Cordova - One of the barriers to scaling Hadoop to 10,000 machines is the single HDFS NameNode. Recent benchmarks show that the HDFS needs to be able to do an order of magnitude more writes per second to reach 10,000. The most promising way to do this is to create a distributed NameNode. Aaron will discuss the issues surrounding distributing the NameNode functionality to multiple machines, including automatically and organically partitioning the namespace, how to keep operations serialized and durable, and how recovery from failure changes. 


Washington, DC  20005 - USA

Tuesday, September 7 at 6:30 PM

Photo: http://photos1.meetupstatic.com/photos/event/a/6/d/event_5702669.jpeg

Attending: 43

Details: http://www.meetup.com/semweb-31/calendar/14521778/