Title :
Dynamic partitioning of non-uniform structured workloads with spacefilling curves
Author :
Pilkington, John R. ; Baden, Scott B.
Author_Institution :
Dept. of Comput. Sci. & Eng., California Univ., San Diego, La Jolla, CA, USA
fDate :
3/1/1996 12:00:00 AM
Abstract :
We discuss inverse spacefilling partitioning (ISP), a partitioning strategy for non-uniform scientific computations running on distributed memory MIMD parallel computers. We consider the case of a dynamic workload distributed on a uniform mesh, and compare ISP against orthogonal recursive bisection (ORE) and a median of medians variant of ORE, ORB-MM. We present two results. First, ISP and ORB-MM are superior to ORE in rendering balanced workloads-because they are more fine-grained-and incur communication overheads that are comparable to ORE. Second, ISP is more attractive than ORB-MM from a software engineering standpoint because it avoids elaborate bookkeeping. Whereas ISP partitionings can be described succinctly as logically contiguous segments of the line, ORB-MM´s partitionings are inherently unstructured. We describe the general d-dimensional ISP algorithm and report empirical results with two- and three-dimensional, non-hierarchical particle methods
Keywords :
distributed memory systems; performance evaluation; resource allocation; software engineering; distributed memory MIMD parallel computers; dynamic partitioning; dynamic workload; inverse spacefilling partitioning; logically contiguous segments; nonuniform scientific computations; nonuniform structured workloads; orthogonal recursive bisection; software engineering; Adaptive mesh refinement; Application software; Computer Society; Computer architecture; Concurrent computing; Costs; Distributed computing; Load management; Partitioning algorithms; Software engineering;
Journal_Title :
Parallel and Distributed Systems, IEEE Transactions on