Hadoop Successes and Failures to Drive Deployment Evolution
3 likes3,086 views
This document summarizes a presentation about successes and failures with Hadoop implementations that have driven its evolution. It discusses initial pseudo-distributed deployments, moving to full clusters, adding monitoring, rack awareness, splitting clusters, retrieving and visualizing data, handling updates, and future directions for Hadoop and related technologies. It emphasizes that Hadoop solutions require ongoing changes and that skills around it are in high demand.
Hadoop Successes and Failures to Drive Deployment Evolution
- 1. Hadoop Hands On Successes and failures to drive evolution Benoit PERROUD Software Engineer @Verisign & Apache Committer GITI BigData, EPFL, November 6. 2012
- 2. Disclaimer • I apologize for speaking “Frenglish” • The views and statements expressed in this talk do not necessarily reflect the views of VeriSign, Inc and any other person involved in the company do not warrant the accuracy, reliability, currency or completeness of those views or statements and do not accept any legal liability whatsoever arising from any reliance on the views, statements and subject matter of the talk. • Apache, Apache Hadoop, Hadoop, Cassandra, Apache Cassandra, Solr, Apache Solr, Hbase, Apache Hbase, Tomcat, Apache Tomcat, Zookeeper, Apache Zookeeper, Lucene, Apache Lucene and the yellow elephant logo are either registered trademarks or trademarks of the Apache Software Foundation in the United States and/or other countries. • Java, Glassfish and the Java logo are registered trademarks of Oracle and/or its affiliates • Python and the Python logo are either registered trademarks or trademarks of the Python Software Foundation • MongoDB, Mongo and the leaf logo are registered trademarks of 10gen, Inc. • All other marks are the property of their respective owners. Verisign Public 2
- 3. Let’s talk about Hadoop! Verisign Public 3
- 4. Hadoop 10k Feet View 1. MapReduce Processing Framework • Map Combine Shuffle Reduce 2. Distributed File System (HDFS) Verisign Public Credit: http://www.michael-noll.com/tutorials/running-hadoop-on-ubuntu-linux-multi-node-cluster/ 4
- 5. Your first Hadoop Deployment • Pseudo-distributed mode on a single node Verisign Public 5
- 6. Going Distributed • TaskTracker (TT) and DataNode (DN) is moved to a dedicated box Verisign Public 6
- 7. NameNode Single Point of Failure • NameNode crashes. Configuring PNN and SNN NFS HA setup is not detailed here. Verisign Public 7
- 8. Bringing Data into the Cluster • Data could be internal to the company, but also external. Data Retrieval and Stream Ingestion are over simplified. Verisign Public 8
- 9. Dealing with API Changes • Integration/Validation Cluster setup Validation Cluster will be omitted in further slides for more clarity Verisign Public 9
- 10. Cluster Is Growing Verisign Public 10
- 12. Turn On Rack Awareness Verisign Public 12
- 13. Split the Cluster to Production and Research Verisign Public 13
- 14. Data Retrieval through REST End Point Verisign Public 14
- 15. Data Retrieval with Search Features Verisign Public 15
- 16. Data Retrieval add Cache Verisign Public 16
- 17. Data Visualization Tools Verisign Public 17
- 18. Upstream Updates Channel Verisign Public 18
- 20. Future Evolutions • Hadoop Next Gen • YARN (2.0) • Graph processing • Neo4J • Google Pregel / Apache Hama • Incremental Updates • Real time ad hoc queries • Cloudera Impala / Google Dremel Verisign Public 20
- 21. Conclusion • Hadoop has gained huge momentum • Technologies (around Hadoop) are evolving really fast • There is no “One size fits all” solution • Design hardly driven by customer needs • Data quality is a hidden requirement Verisign Public 21
- 22. Conclusion #2 • Data Scientists cost a lot • Running on commodity hardware still costs a lot • No one has the full understanding of the full data flow • And you need several FTE just to track the architecture • You have a high risk of misuse of these softwares • Hiring engineers with deep knowledge (meaning: hands on experience) in some of these softwares is already a challenge Verisign Public 22
- 23. Recommended Reading Hadoop In Practice by Alex Holmes Senior Software Engineer @Verisign Verisign Public 23
- 24. Q&A Benoit PERROUD bperroud@verisign.com Verisign Public 24
- 25. Thank You © 2012 VeriSign, Inc. All rights reserved. VERISIGN and other trademarks, service marks, and designs are registered or unregistered trademarks of VeriSign, Inc. and its subsidiaries in the United States and in foreign countries. All other trademarks are property of their respective owners.