nceskurachi / taskdep

Task graph generator

Introduction

This tool generates task graphs, modeling precedence constraints among tasks and communication requirements.

A task is formally defined by:

• edges, representing tasks. A specific worst-case execution time features each task
• vertices, representing precedence constraints among tasks. Each of them is characterized by the amount of exchanged data. Such dependencies are represented as: A -> B. Meaning that the task B must start not earlier than the end of task A.

The software implements two different generators.

The first algorithm is called CODESTYLE generator and basically does:

1. divide the actual task set into two groups
2. decide randomly how to execute the two groups: in parallel or sequentially
3. repeat recursively step 1 for the two groups See the last section for two examples.

The execution flow of the second algorithm (DATA-FLOW generator) is the following:

1. generate the task set
2. pick a task from the task set
3. pick a second task from the neighborhood of the first task
4. create a dependency between them
5. repeat step 2 until the number of desired dependencies is reached

Required packages

Required libraries:

• GFlags Library: http://code.google.com/p/gflags/
• Protobuf Library: http://code.google.com/p/protobuf/

Suggested software:

• Graphviz: http://www.graphviz.org/

How to

Compile:

make

Once the software has been compiled, execute:

./software --binaryFile BINARY_OUTPUT_FILE [OPTIONS]

where:

• binaryFile: Binary file with the generated task graph (string)
• plainTextFile: Plain text with the generated task graph (string)
• graphvizFile: DOT file with the generated task graph (string)
• maxWcet: Maximum Worst-Case Execution Time (int32)
• minWcet: Minimum Worst-Case Execution Time (int32)
• maxExchangedData: Maximum amount of exchanged data (int32)
• minExchangedData: Minimum amount of exchanged data (int32)
• dependencyDistance: Maximum dependency distance (int32)
• nOfDependencies: Number of Dependencies (int32)
• nOfTasks (Number of tasks) type: int32 default: 15
• probSeqPar (Ser/Par probability [0, 1]) type: double default: 0.5
• taskGenerator: Task generator to use:
  • 0: code-style generator (SUGGESTED ALGORITHM)
  • 1: data-flow generator

A script has been provided to make the use easier. It automatically generate the binary file and the png image.

./script.sh [GENERATIONNAME]

Suppose you type:

./script example

then, the following files are generated

• example.bin: serialized tash graph (using the Protobuf library)
• example.txt: plain text task graph description
• example.dot: task graph representation using DOT language
• example.png: graphical representation of the task graph (using dot command)

In case you want to regenerate source files for the serialization:

make gen
make

Other software

Alternative software that you would like to test:

• TGFF: http://ziyang.eecs.umich.edu/~dickrp/tgff/
• STG: http://www.kasahara.elec.waseda.ac.jp/schedule/
• TaskGraphGenerator: http://sourceforge.net/projects/taskgraphgen/
• http://ai.unibo.it/node/410

Example

Let us consider two examples, obtained by using the CODE-STYLE algorithm.

The first one contains 30 tasks (without considering synchronization points) with a probability 0.5 (within [0.0, 1.0]) of having either a parallel or sequential execution.

The second example contains 10 tasks and probability 0.7 (parallel executions are more probable than sequential ones).