CALL FOR PAPERS
Special Issue of the Journal of Parallel and Distributed Computing on Dynamic Load Balancing
Papers are solicited for a special issue of the Journal of Parallel and Distributed Computing (JPDC) to be tentatively scheduled for publication in January 1988. Dynamic load balancing is extremely important for efficient adaptive grid-based computations on distributed-memory multiprocessors. Numerical solutions to problems in computational fluid dynamics, computational chemistry, and semiconductor simulation involve the use of structured and/or unstructured grids. Parallel computation of such problems on distributed-memory machines typically involves two steps. First, as a preprocessing step, the computational mesh is partitioned into smaller submeshes. Second, the partitioned submeshes are assigned to processors based on a mapping strategy. While this static partitioning and mapping approach is appropriate for steady-state calculations that do not change in computational intensity over time and space, it is grossly inefficient for unsteady, adaptive calculations.
As the computational behavior changes nonuniformly, some processors may have much more work to perform than others. The imbalance in the computational load implies that the initial partitioning of the mesh is no longer efficient. It is therefore indispensable that the amount of work each processor performs be balanced at runtime in order to increase processor utilization and improve performance. Balancing the runtime computational load, however, is often very difficult due to many practical issues. These include a reliable measurement of the computational load, the amount of runtime data movement, and the minimization of interprocessor communication. The main purpose of this special issue is to report on state-of-the-art dynamic load balancing methods for real-world computational science and engineering problems.
Topics of interest include, but are not limited to, frameworks for global and/or local load balancing, comparative studies of load balancing methods, fast and high quality mesh partitioners, comparative studies of mesh partitioners, metrics for accurate load measurement, models to predict communication behavior relevant to load balancing, discussions of the effects of load balancing on real-world problems, runtime data/workload distribution and migration methods to achieve load balancing, and parallel mesh adaption procedures. Authors should follow the JPDC manuscript format as described in the "Information for Authors" at the end of each issue of JPDC.