Computer Science and Engineering, Department of



Harkamal Walia

Date of this Version



2014 IEEE 28th International Parallel & Distributed Processing Symposium Workshops, Pages: 546 - 555, DOI: 10.1109/IPDPSW.2014.66 IEEE Conference Publications.


Copyright © 2014 IEEE. Used by permission.


Scientific workflows are a useful tool for managing large and complex computational tasks. Due to its intensive resource requirements, the scientific workflows are often executed on distributed platforms, including campus clusters, grids and clouds. In this paper we build a scientific workflow for blast2cap3, the protein-guided assembly, using the Pegasus Workflow Management System (Pegasus WMS). The modularity of blast2cap3 allows us to decompose the existing serial approach on multiple tasks, some of which can be run in parallel. Afterwards, this workflow is deployed on two distributed execution platforms: Sandhills, the University of Nebraska Campus Cluster, and the Open Science Grid (OSG). We compare and evaluate the performance of the built workflow for the both platforms. Furthermore, we also investigate the influence of the number of clusters of transcripts in the blast2cap3 workflow over the total running time. The performed experiments show that the Pegasus WMS implementation of blast2cap3 significantly reduces the running time compared to the current serial implementation of blast2cap3 for more than 95 %. Although OSG provides more computational resources than Sandhills, our workflow experimental runs have better running time on Sandhills. Moreover, the selection of 300 clusters of transcripts gives the optimum performance with the resources allocated from Sandhills.