Hadoop For Administrators

• comfortable with basic Linux system administration
• basic scripting skills

Knowledge of Hadoop and Distributed Computing is not required, but will be introduced and explained in the course.

Lab environment

Zero Install : There is no need to install hadoop software on students’ machines! A working hadoop cluster will be provided for students.

Students will need the following

• an SSH client (Linux and Mac already have ssh clients, for Windows Putty is recommended)
• a browser to access the cluster. We recommend Firefox browser with FoxyProxy extension installed

Apache Hadoop is the most popular framework for processing Big Data on clusters of servers. In this three (optionally, four) days course, attendees will learn about the business benefits and use cases for Hadoop and its ecosystem, how to plan cluster deployment and growth, how to install, maintain, monitor, troubleshoot and optimize Hadoop. They will also practice cluster bulk data load, get familiar with various Hadoop distributions, and practice installing and managing Hadoop ecosystem tools. The course finishes off with discussion of securing cluster with Kerberos.

“…The materials were very well prepared and covered thoroughly. The Lab was very helpful and well organized”
— Andrew Nguyen, Principal Integration DW Engineer, Microsoft Online Advertising

Audience

Hadoop administrators

Format

Lectures and hands-on labs, approximate balance 60% lectures, 40% labs.

• Introduction
  • Hadoop history, concepts
  • Ecosystem
  • Distributions
  • High level architecture
  • Hadoop myths
  • Hadoop challenges (hardware / software)
  • Labs: discuss your Big Data projects and problems
• Planning and installation
  • Selecting software, Hadoop distributions
  • Sizing the cluster, planning for growth
  • Selecting hardware and network
  • Rack topology
  • Installation
  • Multi-tenancy
  • Directory structure, logs
  • Benchmarking
  • Labs: cluster install, run performance benchmarks
• HDFS operations
  • Concepts (horizontal scaling, replication, data locality, rack awareness)
  • Nodes and daemons (NameNode, Secondary NameNode, HA Standby NameNode, DataNode)
  • Health monitoring
  • Command-line and browser-based administration
  • Adding storage, replacing defective drives
  • Labs: getting familiar with HDFS command lines
• Data ingestion
  • Flume for logs and other data ingestion into HDFS
  • Sqoop for importing from SQL databases to HDFS, as well as exporting back to SQL
  • Hadoop data warehousing with Hive
  • Copying data between clusters (distcp)
  • Using S3 as complementary to HDFS
  • Data ingestion best practices and architectures
  • Labs: setting up and using Flume, the same for Sqoop
• MapReduce operations and administration
  • Parallel computing before mapreduce: compare HPC vs Hadoop administration
  • MapReduce cluster loads
  • Nodes and Daemons (JobTracker, TaskTracker)
  • MapReduce UI walk through
  • Mapreduce configuration
  • Job config
  • Optimizing MapReduce
  • Fool-proofing MR: what to tell your programmers
  • Labs: running MapReduce examples
• YARN: new architecture and new capabilities
  • YARN design goals and implementation architecture
  • New actors: ResourceManager, NodeManager, Application Master
  • Installing YARN
  • Job scheduling under YARN
  • Labs: investigate job scheduling
• Advanced topics
  • Hardware monitoring
  • Cluster monitoring
  • Adding and removing servers, upgrading Hadoop
  • Backup, recovery and business continuity planning
  • Oozie job workflows
  • Hadoop high availability (HA)
  • Hadoop Federation
  • Securing your cluster with Kerberos
  • Labs: set up monitoring
• Optional tracks
  • Cloudera Manager for cluster administration, monitoring, and routine tasks; installation, use. In this track, all exercises and labs are performed within the Cloudera distribution environment (CDH5)
  • Ambari for cluster administration, monitoring, and routine tasks; installation, use. In this track, all exercises and labs are performed within the Ambari cluster manager and Hortonworks Data Platform (HDP 2.0)