A higher order parallel abstraction for fixed degree divide and conquer problem solving strategy

In this paper, we present an algorithmic framework for the Fixed Degree Divide & Conquer(FDDC) problem solving strategy, that adapts the divide and conquer algorithm for parallel computations. We capture the notion of a thread of execution in an active task object which are instances of a data type called the tasktype and propose a higher order, tasking abstraction called maptask that takes as its arguments, a tasktype, the (fixed) degree of the runtime task hierarchy and the arguments for the tasks. We have implemented the above abstraction by extending C with task-type and maptask in a package called the CT Kernel that currently executes tasks in a distributed environment consisting of 10 netwoked VAXstations. Using the above abstractions we have implemented the FDDC algorithms for mergesort, FFT, Cubic spline based curve fitfing, Fractals, Matrix multiplication, as well as Iterative relaxation.