SocialGen

SocialGen is a tool for generating large quantities of synthetic data that models status updates and sent messages from users of two imaginary social networks: Gleambook and Chirp networks. SocialGen can be configured to run in a parallel/distributed fashion generating partitions of data in parallel using a cluster of machines.

SocialGen can generate data both in Asterix Data Model (ADM) format and JSON. It can also generate data with either 64-bit integer identifiers or string identifiers. Here is a sample data type definition for the output of SocialGen using 64-bit integer identifiers.

create type EmploymentType as {
  organization: string,
  start_date: date,
  end_date: date?
}

create type GleambookUserType as {
  id: int64,
  alias: string,
  name: string,
  user_since: datetime,
  friend_ids: {{ int64 }},
  employment: [EmploymentType]
}

create type GleambookMessageType as {
  message_id: int64,
  author_id: int64,
  in_response_to: int64?,
  sender_location: point,
  send_time: datetime,
  message: string
}

create type ChirpUserType as {
  screen_name: string,
  lang: string,
  friends_count: int32,
  statuses_count: int32,
  name: string,
  followers_count: int32
}

create type ChirpMessageType as {
  chirpid: int64,    
  user: ChirpUserType,
  sender_location: point,
  send_time: datetime,
  referred_topics: {{ string }},
  message_text: string
}

When using string identifiers, int64 needs to be replaced with string in this type definition.

Using SocialGen

Prerequisites

A suitable *nix environment (Linux, OSX)
JDK 1.8+
Maven 3.1.1 or greater

Steps

Check out the SocialGen project in a directory via git. Assume that the path to the directory is $HOME/SocialGen. We will refer to this directory as SOCIAL_GEN_HOME in the rest of this document. Important If you intend to generate data in parallel using a cluster of machines, it is required that SOCIAL_GEN_HOME is on the NFS (Network File System) and accessible from all the participating machines.
Set SOCIAL_GEN_HOME as an environment variable on the machine you are running SocialGen from:

> export SOCIAL_GEN_HOME=$HOME/SocialGen

Go into the SOCIAL_GEN_HOME and build its artifacts by executing the following commands:

> cd $SOCIAL_GEN_HOME
> mvn clean package

Upon a successful build, a new directory named target will be created under SOCIAL_GEN_HOME that contains the jar file for SocialGen i.e. SocialGen.jar .

The configuration files for SocialGen are under the conf directory:
machines: This file lists the machine(s) that you want to generate social data on. If you want to generate data only on a single machine, this file may contain the hostname for that machine or simply localhost as the machine name (if everything is done locally). For generating data on a cluster, you need to add the machine names (one machine name per line) into the machines file. Important: SocialGen assumes that the machines that are listed in this file are accessible via password-less SSH.
conf.xml: This file describes the input parameters to SocialGen and properties of each partition of the generated data. The random data generation process in SocialGen uses a single global seed to control the generation process. This seed is specified in the conf.xml file. In a multi-partition case, SocialGen assigns a unique seed, derived from the global seed, to each partition. In the conf.xml file, you also need to specify the total number of GleambookUser and ChirpUser records in the social network along with the average number of messages per user. Moreover, for each partition of the generated data, you specify a unique partition id (Important: The partition id has to be a unique numeric value (64-bit integer) per partition), the machine that the partition resides on and the absolute path for that partition. Please note that the machine should be a valid and accessible machine that is already listed in the machines file (see above) and the path to the partition should be writable.

As an example, imagine we have two machines: rainbow-1 and rainbow-2, while on each one of them we want to have two partitions of generated data such that partitions 1 and 2 reside on rainbow-1 and partitions 3 and 4 reside on rainbow-2. Moreover, assume we decide to have 1000 users in the GleambookUsers dataset with an average of 5 messages per user and 2000 users in the ChirpUsers with an average of 10 chirp messages per user. The configuration files for this example would look like as:

machines file:

rainbow-1
rainbow-2

conf.xml file:

<?xml version="1.0" encoding="UTF-8"?>
<Dbgen>
	<gleambook.users.count>1000</gleambook.users.count>
  <chirp.users.count>1000</chirp.users.count>
  <avg.message.count.per.gleambook.user>5</avg.message.count.per.gleambook.user>
  <avg.message.count.per.chirp.user>10</avg.message.count.per.chirp.user>
  <seed>10</seed>

	<partitions>
		<partition>
			<id>1</id>
			<host>rainbow-1</host>
			<path>/mnt/data/sda/socialData</path>
		</partition>

		<partition>
			<id>2</id>
			<host>rainbow-1</host>
			<path>/mnt/data/sdb/socialData</path>
		</partition>

		<partition>
			<id>3</id>
			<host>rainbow-2</host>
			<path>/mnt/data/sda/socialData</path>
		</partition>

		<partition>
			<id>4</id>
			<host>rainbow-2</host>
			<path>/mnt/data/sdb/socialData</path>
		</partition>
 
	</partitions>
</Dbgen>

SocialGen needs to distribute necessary artifacts and scripts to the participating machines. For simplicity, we choose a single path which needs to be available locally on all the machines. SocialGen creates this path and its corresponding directory on the local file system of each machine and copies the necessary artifacts to these directories. We refer to this directory path as SOCIAL_GEN_LOCAL_HOME. You need to set this path as an environment variable (on the machine you are running socialGen from):

> export SOCIAL_GEN_LOCAL_HOME=<absolute path to the chosen directory>

Install SocialGen by running the install script. This step copies the required artifacts from SOCIAL_GEN_HOME into SOCIAL_GEN_LOCAL_HOME on all machines:

> $SOCIAL_GEN_HOME/scripts/install.sh

Run SocialGen to initialize the data generation process:

For generating data in ADM format with 64-bit integer identifiers, you simply run:

> $SOCIAL_GEN_HOME/scripts/init.sh

For generating data in JSON format with string identifiers, you need to invoke the init.sh script as:

 > $SOCIAL_GEN_HOME/scripts/init.sh -f JSON -k STRING

Upon termination of the generation process, you can find three data files (in .adm or .json format, depending on your preference (see previous step)) per partition under the path that is specified for the partition in the conf.xml file.

parshimers/socialGen

SocialGen

Using SocialGen

Prerequisites

Steps