Testbed

This project is used to execute multiple experiments to compare Spark vs MapReduce. Before creating a one-JAR with all dependencies, make sure the following dependencies are installed:

• Maven
• Java JDK 11

Then, run the following script:

$ bash scripts/install.sh

Afterwards, it will create a target folder, which holds the one-JAR with all dependencies.

To run the JAR, use the following command:

$ java -jar Testbed-1.0-SNAPSHOT-jar-with-dependencies.jar [options]

Where [options] are the following:

usage: java -jar Testbed-1.0-SNAPSHOT-jar-with-dependencies.jar [options]
 -f,--framework-name <arg>               Data Processing Framework's name.
                                         Available options are:
                                         [MapReduce, Spark]
 -i,--instrumented                       If this flag is present, the
                                         Testbed will use the instrumented
                                         invocations. Without this flag,
                                         the Testbed will use the
                                         invocations required to measure
                                         time
 -l,--local                              If the flag is present, the
                                         Testbed uses the local
                                         environment for the frameworks.
                                         Without this flag, the Testbed
                                         uses the cluster environment
 -o,--output <arg>                       Output file path
 -p,--pipeline <arg>                     Pipeline file path
 -s,--sheet-name <arg>                   Sheet name in output file path
 -t,--tolerable-error-percentage <arg>   Tolerable error percentage.
                                         Default value is: 5.0

This section is extracted from usage, which can always be displayed when invoking the java JAR with no options, as follows:

$ java -jar Testbed-1.0-SNAPSHOT-jar-with-dependencies.jar

The pipeline refers to a JSON file with the following aspect:

[
  {
    "operation": "LOAD",
    "outputTag": "dataset1",
    "datasetDirectoryPath": "input/Ad_click_on_taobao_10000"
  },
  {
    "operation": "SELECT",
    "inputTag": "dataset1",
    "outputTag": "selectedDataset1",
    "rowsSelectivityFactor": 0.005,
    "columnName": "DateTime"
  }
]

An example of invocation with inputParameters is the following:

$ java -jar Testbed-1.0-SNAPSHOT-jar-with-dependencies.jar \
--tolerable-error-percentage 5.0 \
--framework-name MapReduce \
--instrumented \
--pipeline pipelines/pipeline.json \
--output output/operation_instrumentations.xlsx \
--sheet-name Dataset_5% \
--local

To execute experiments, we have created the script scripts/experiments_runner.sh. This script executes each experiment for Spark and MapReduce 3 times and an additional time to instrument the execution. The individual output of each experiment gets aggregated into a single output file. To execute this script, use the following command:

$ bash scripts/experiments_runner.sh

For more information, check the dissertation associated to this project.