Awesome
Information
SLEIPNIR is a DAG scheduling simulator focused on mobile cloud/edge/iot infrastructures released under MIT license. If you use it in your papers, please cite my work using the following bibtex entry:
@inproceedings{DBLP:conf/ccgrid/MaioB18,
author = {Vincenzo De Maio and
Ivona Brandic},
editor = {Esam El{-}Araby and
Dhabaleswar K. Panda and
Sandra Gesing and
Amy W. Apon and
Volodymyr V. Kindratenko and
Massimo Cafaro and
Alfredo Cuzzocrea},
title = {First Hop Mobile Offloading of {DAG} Computations},
booktitle = {18th {IEEE/ACM} International Symposium on Cluster, Cloud and Grid
Computing, {CCGRID} 2018, Washington, DC, USA, May 1-4, 2018},
pages = {83--92},
publisher = {{IEEE} Computer Society},
year = {2018},
url = {https://doi.org/10.1109/CCGRID.2018.00023},
doi = {10.1109/CCGRID.2018.00023},
timestamp = {Wed, 16 Oct 2019 14:14:53 +0200},
biburl = {https://dblp.org/rec/conf/ccgrid/MaioB18.bib},
bibsource = {dblp computer science bibliography, https://dblp.org}
}
Prerequisites
SLEIPNIR is written in Java, therefore the latest version of JDK is required. Dependencies are resolved using Apache Maven, which is available at: https://maven.apache.org/download.cgi Also, SLEIPNIR uses Apache Spark to distribute load on multiple cores. Apache Spark can be downloaded at: https://spark.apache.org/downloads.html
Download and install SLEIPNIR
The simulator is available from github (https://github.com/vindem/sleipnir)
To download the latest version, you need to clone the repository running
git clone https://github.com/vindem/sleipnir
To install the simulator, you need to run
mvn clean package -Dmaven.test.skip
Which generates the jar archive needed by Apache Spark. In alternative, you can also use the available build.sh script
./build.sh
This command will generate the jar sleipnir.jar in the subfolder target/.
Running simulation with SLEIPNIR
“HelloWorld” example
In the HelloWorld example, we simulate offloading of a workflow composed of 5 sequential FACEBOOK DAGs, executed by 1000 mobile devices. Computational infrastructure is composed of 6 cloud nodes. We simulate mobility over the area of HERNALS, divided in hexagonal cells of 2 km radius with an edge node in each cell. For offloading, HEFT list-based algorithm is used. The “HelloWorld” example is described in the source file OffloadingHelloWorld.java.
Running HelloWorld example
To run SLEIPNIR, just run
spark-submit target/sleipnir.jar
Simulation Setup
Simulation can be configured either by using command lines arguments or the configuration file simulation.json. The arguments that can be used are the following:
- -h, -? Prints usage information
- -mobile=n Instantiates n mobile devices
- -cloud=n Instantiates n cloud nodes
- -wlRuns=n Each workflows has n applications
- -cloudonly=true/false If true, simulation uses only Cloud nodes
- -area=name Urban area where the offloading is performed (possible choices: HERNALS, LEOPOLDSTADT, SIMMERING)
- -eta=n Sets the eta parameter, which is necessary to set offloading cost (the higher the eta, the lower the cost).
- -outfile=string Saves output in file filename
- -iter=n Executes simulation for n iterations
- -navigatorMapSize=# Lambda parameter for size of navigator MAP (in kb)
- -antivirusFileSize=# Lambda parameter for size of antivirus file (in kb)
- -facerecImageSize=# Lambda parameter for size of image file (in kb) for Facerec app
- -chessMi=# Lambda parameter for computational size of Chess app
- -navigatorDistr=# Probability of NAVIGATOR app in workflow (must be between 0 and 1).
- -antivirusDistr=# Probability of ANTIVIRUS app in workflow (must be between 0 and 1).
- -facerecDistr=# Probability of FACEREC app in workflow (must be between 0 and 1).
- -chessDistr=# Probability of CHESS app in workflow (must be between 0 and 1).
- -facebookDistr=# Probability of FACEBOOK app in workflow (must be between 0 and 1).
- -mobility=true/false If true, SLEIPNIR simulates mobility of users using sumo trace files. Example files for the areas of HERNALS, LEOPOLDSTADT and SIMMERING are available here