Innovations and Interdisciplinary Solutions for Underserved Areas. Second International Conference, InterSol 2018, Kigali, Rwanda, March 24–25, 2018, Proceedings

Research Article

BEDWE: A Decentralized Workflow Engine for Best-Effort Infrastructures

Download
180 downloads
  • @INPROCEEDINGS{10.1007/978-3-319-98878-8_22,
        author={Palakiyem Wallah and C\^{e}dric Tedeschi and Jean-Louis Pazat},
        title={BEDWE: A Decentralized Workflow Engine for Best-Effort Infrastructures},
        proceedings={Innovations and Interdisciplinary Solutions for Underserved Areas. Second International Conference, InterSol 2018, Kigali, Rwanda, March 24--25, 2018, Proceedings},
        proceedings_a={INTERSOL},
        year={2018},
        month={9},
        keywords={Workflows Decentralized orchestration Fault-tolerance Best-effort infrastructures},
        doi={10.1007/978-3-319-98878-8_22}
    }
    
  • Palakiyem Wallah
    Cédric Tedeschi
    Jean-Louis Pazat
    Year: 2018
    BEDWE: A Decentralized Workflow Engine for Best-Effort Infrastructures
    INTERSOL
    Springer
    DOI: 10.1007/978-3-319-98878-8_22
Palakiyem Wallah1, Cédric Tedeschi2,*, Jean-Louis Pazat2
  • 1: Université de Kara
  • 2: Univ Rennes, Inria, CNRS, IRISA
*Contact email: cedric.tedeschi@inria.fr

Abstract

We consider the problem of executing composite computing applications called on top of unreliable computing infrastructures. Having in mind the situation of the electric delivery in the sub-saharan area, we propose BEDWE, a decentralized workflow engine able to dynamically assign portions of the workflow to currently live compute nodes. More precisely, in a point-to-point manner, each node can receive a part of the workflow and delegate a subpart of it to another node. This mechanism can be repeated recursively until the whole workflow is executed. BEDWE includes a mechanism to support nodes leaving the network due to power outage. We present a software prototype of BEDWE and its experimentation over the French nation-wide Grid’5000 platform.