Title: Parallelisation of a watershed distributed ecohydrological model with dynamic task scheduling

Authors: Lajiao Chen; Yan Ma; Peng Liu; Wei Xue

Addresses: Institute of Remote Sensing and Digital Earth, Chinese Academy of Sciences, Beijing 100094, China ' Institute of Remote Sensing and Digital Earth, Chinese Academy of Sciences, Beijing 100094, China ' Institute of Remote Sensing and Digital Earth, Chinese Academy of Sciences, Beijing 100094, China ' Department of Computer Science and Engineering, Tsinghua University, Beijing 100084, China

Abstract: Watershed distributed ecohydrological modelling associating with massive data and intensive computation, has a rising demand for performance computing. Till now models parallelisation mainly conducted at a granularity of sub-basin, which is of low parallel efficiency and tends to cause load unbalance. Few studies conducted at a granularity of grid cell, which has very complicated control logic, leading to poor stability and extensibility of parallel computing. This paper presents a dynamic task scheduling based parallelism scheme with fine granularity and simple control logic. The basic concept is to decouple the dependence among grid-based tasks according to flow connectivity. Independent tasks are parallel executed while dependent tasks are dynamic proceed based on a dynamic task-tree scheduling solution. The approach is applied in an experimental watershed and results illustrate that this approach can highly promote computing speed. Therefore, the approach proposed in this study is valuable for watershed distributed ecohydrological models application.

Keywords: watershed modelling; ecohydrological modelling; distributed models; ecohydrology; parallelisation; dynamic task scheduling; DAG model; fine granularity; control logic; modelling; flow connectivity; hydrology.

DOI: 10.1504/IJAHUC.2014.065774

International Journal of Ad Hoc and Ubiquitous Computing, 2014 Vol.17 No.2/3, pp.110 - 121

Received: 10 Jul 2013
Accepted: 14 Oct 2013

Published online: 19 Nov 2014 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article