DDOps: Dual-direction operations for load balancing on non-dedicated heterogeneous distributed systems

Nader Mohamed, Jameela Al-Jaroodi, Hong Jiang

    Research output: Contribution to journalArticlepeer-review

    9 Citations (Scopus)

    Abstract

    Given the increasing prevalence of compute/data intensive applications, the explosive growth in data, and the emergence of cloud computing, there is an urgent need for effective approaches to support such applications in non-dedicated heterogeneous distributed environments. This paper proposes an efficient technique for handling parallel tasks, while dynamically maintaining load balancing. Such tasks include concurrently downloading files from replicated sources; simultaneously using multiple network interfaces for message transfers; and executing parallel computations on independent distributed processors. This technique, DDOps, (Dual Direction Operations) enables efficient utilization of available resources in a parallel/distributed environment without imposing any significant control overhead. The idea is based on the notion of producer pairs that perform tasks in parallel from opposite directions and the consumers that distribute and control the work and receive and combine the results. Most dynamic load balancing approaches require prior knowledge and/or constant monitoring at run time. In DDOps, load balancing does not require prior knowledge or run-time monitoring. Rather, load balancing is automatically inherent as the tasks are handled from the opposite directions, allowing the processing to continue until the producers meet indicating the completion of all tasks at the same time. Thus DDOps is most suitable for heterogeneous environments where resources vary in specifications, locations, and operating conditions. In addition, since DDOps does not require producers to communicate at all, the network effect is minimized.

    Original languageEnglish
    Pages (from-to)503-528
    Number of pages26
    JournalCluster Computing
    Volume17
    Issue number2
    DOIs
    Publication statusPublished - Jan 2014

    Keywords

    • Distributed systems
    • Dual-direction processing
    • Heterogeneous resources
    • Load balancing
    • Parallel computing

    ASJC Scopus subject areas

    • Software
    • Computer Networks and Communications

    Fingerprint

    Dive into the research topics of 'DDOps: Dual-direction operations for load balancing on non-dedicated heterogeneous distributed systems'. Together they form a unique fingerprint.

    Cite this