This is a set of scripts used for deploying a GTStream benchmark on an OpenStack cloud.

Currently these are highly personalized to my settings, but should be easy to extend/reconfigure.

Author: Brian Laub - [email protected]

Known Issues

Before using, be aware of some known issues (mostly due to my laziness):

• the user name and path "/home/ubuntu" is hardcoded all over the place :(
• several other paths are hardcoded (e.g. looking in "~/files" for required JVM package)
• OpenStack dependencies: this has never been tested (nor do I have any intention to in the near future) on AWS.

Intro

GTStream (https://github.com/chengweiwang/GTStream) is designed as a large-scale log-streaming benchmark. This project adds some scripts that attempt to make it easier to deploy a benchmark test on an OpenStack cloud.

Currently, these scripts use newer versions of software used by GTStream, including:

• Hadoop 1.2.1
• HBase 0.94.10
• Flume 1.4.0

Getting Started

Deploying requires Hadoop, HBase, and Flume. You can set these up in any configuration you like, but the scripts included here attempt to make it a bit easier to scale up a cloud deployment of these packages.

To get started, first edit the following two files to adjust settings:

• conf/cloud-properties.sh: set certain VM properties to use, such as ssh keyname, path to ssh private key, login user, VM image, security groups, etc.
• conf/gtstream-env.sh: environment variables used by install scripts. Set things like software versions, URLs to download packages from, install paths, configuration files, etc.

You can build up a minimal set of "required" servers by running:

bin/bootup_required_nodes.sh

which will start:

• a Hadoop Master
• an HBase Master
• a Zookeeper quorum (at least one server)

Installing Software on Instances

Once an instance has been booted, you need to copy a JVM to "~/files" on that instance, and you need a copy of these scripts. Two included scripts can be used to do this. Here's a short example:

1. download an Oracle JVM to a directory called "files" somewhere. Note that you may not be able to use wget from the Oracle website due to licensing stuff: mkdir /tmp/files && cd /tmp/files wget [some_jvm_url]
2. clone the scripts: git clone [this_repos_url] /tmp/gtstream-scripts && cd /tmp/gtstream-scripts
3. bootup some nodes (see "Getting Started" above): bin/bootup_required_nodes.sh - or - bin/bootup_node_and_wait.sh [flavor] [hostname]
4. copy JVM and scripts: bin/sync_files_to_all.sh bin/sync_scripts_to_all.sh

Once you've got it all sync'd up, ssh into your VM and run: cd gtstream-scripts sudo bin/install.sh

Starting Software

To start Hadoop on the master, run: start-all.sh Note you may want to remove localhost from HADOOP_HOME/conf/slaves if you don't want the master to run a tasktracker/datanode.

To start HBase on the master, run: start-hbase.sh Note you probably don't want to remove localhost from HBASE_HOME/conf/regionservers, because localhost will likely be the regionserver for -ROOT-.

Scaling

To add Hadoop slaves (a datanode and a tasktracker), bootup a VM using bin/bootup_node_and_wait.sh and follow the install procedure above. Then login and run: bin/add_self_to_hadoop.sh

To add HBase regionservers, do the same and login and run: bin/add_self_to_hbase.sh

GTStream Benchmarking

By default, Flume is configured to tail a single logfile using the exec source, and to push data directly into HBase.

You can (and probably will want to) customize this to a more intricate setup (e.g., with log aggregation, and using something more stable than exec source).