Design and Implementation of Kepler Workflows for BioEarth

详细信息    查看全文

• 作者：Tristan Mullis^a ; Mingliang Liu^b ; Ananth Kalyanaraman^a ; ^{ananth@eecs.wsu.edu" class="auth_mail} ; Joseph Vaughan^b ; Christina Tague^c ; Jennifer Adam^b
• 关键词：Kepler ; scientific workflows ; environmental modeling ; Earth system modeling ; data-intensive application ; compute-intensive application ; parallel workflows.
• 刊名：Procedia Computer Science
• 出版年：2014
• 年：2014
• 卷：29
• 期：Complete
• 页码：1722-1732
• 全文大小：971 K

文摘

BioEarth is an ongoing research initiative for the development of a regional-scale erline">Earth erline">System erline">Model (EaSM) for the U.S. Pacific Northwest. Our project seeks to couple and integrate multiple stand-alone EaSMs developed through independent efforts for capturing natural and human processes in various realms of the biosphere: atmosphere (weather and air quality), terrestrial biota (crop, rangeland, and forest agro-ecosystems) and aquatic (river flows, water quality, and reservoirs); hydrology links all these realms. Due to the need to manage numerous complex simulations,an application of automated workflows was essential. In this paper, we present a case study of workflow design for the BioEarth project using the Kepler system to manage applications of the erline">Regional erline">Hydro-erline">Ecologic erline">Simulation erline">System (RHESSys) model. In particular, we report on the design of Kepler workflows to support: 1) standalone executions of the RHESSys model under serial and parallel applications, and 2) a more complex case of performing calibration runs involving multiple preprocessing modules, iterative exploration of parameters and parallel RHESSys executions. We exploited various Kepler features including a user-friendly design interface and support for parallel execution on a cluster. Our experiments show a performance speedup between 7–12x, using 16 cores of a Linux cluster, and demonstrate the general effectiveness of our Kepler workflows in managing RHESSys runs. This study shows the potential of Kepler to serve as the primary integration platform for the BioEarth project, with implications for other data- and compute-intensive Earth systems modeling projects.