5th International ICST Conference on Collaborative Computing: Networking, Applications, Worksharing

Research Article

Flexible failure handling for cooperative processes in distributed systems

Download72 downloads
  • @INPROCEEDINGS{10.4108/ICST.COLLABORATECOM2009.8306 ,
        author={Artin Avanes and Johann-Christoph Freytag},
        title={Flexible failure handling for cooperative processes in distributed systems},
        proceedings={5th International ICST Conference on Collaborative Computing: Networking, Applications, Worksharing},
        proceedings_a={COLLABORATECOM},
        year={2009},
        month={12},
        keywords={Computer networks Concurrent computing Distributed computing Handheld computers Parallel processing Performance gain Protocols Prototypes Sensor systems Wireless sensor networks},
        doi={10.4108/ICST.COLLABORATECOM2009.8306 }
    }
    
  • Artin Avanes
    Johann-Christoph Freytag
    Year: 2009
    Flexible failure handling for cooperative processes in distributed systems
    COLLABORATECOM
    ICST
    DOI: 10.4108/ICST.COLLABORATECOM2009.8306
Artin Avanes1,*, Johann-Christoph Freytag1,*
  • 1: Database and Information System Group, Humboldt-Universität zu Berlin, Unter den Linden 6, 10099 Berlin, Germany
*Contact email: avanes@informatik.hu-berlin.de, freytag@informatik.hu-berlin.de

Abstract

Distributed systems will be increasingly built on top of wireless networks, such as sensor networks or hand-held devices with advanced sensing and computational abilities. Supporting cooperative processes executed by such unreliable and dynamic system components poses a various number of new technical challenges. In terms of recovery, limited resource capabilities have be considered during re-scheduling of failed process activities. In terms of concurrency, a non-blocking protocol is required to allow a high degree of parallelism. In this paper, we introduce a flexible and resource-oriented failure handling mechanism for cooperative processes in hierarchical and distributed systems. The objective is to ensure both - transactional semantics as well as the selection of suitable nodes with respect to available resource capabilities. Based on a nested execution model, we develop a multi-stage algorithm that uses constraint solving techniques in a parallel fashion thus achieving a more efficient recovery. We evaluate our proposed techniques in a prototype implementation, and demonstrate significant performance gains by using a parallel re-scheduling.