Title: Applicability of the software cost model COCOMO II to HPC projects

Authors: Julian Miller; Sandra Wienke; Michael Schlottke-Lakemper; Matthias Meinke; Matthias S. Müller

Addresses: IT Center, RWTH Aachen University, Aachen, Germany; JARA – High-Performance Computing, Aachen, Germany ' IT Center, RWTH Aachen University, Aachen, Germany; JARA – High-Performance Computing, Aachen, Germany ' Institute of Aerodynamics, RWTH Aachen University, Aachen, Germany; JARA – High-Performance Computing, Aachen, Germany ' Institute of Aerodynamics, RWTH Aachen University, Aachen, Germany ' IT Center, RWTH Aachen University, Aachen, Germany; JARA – High-Performance Computing, Aachen, Germany

Abstract: The complexity of parallel computer architectures continuously increases with the pursuit of exaflop computing, which makes accurate development effort estimation and modelling more important than ever. While sophisticated cost models are widely used in traditional software engineering, they have rarely been investigated for the performance-oriented HPC domain. Therefore, we evaluate the fit and accuracy of the popular COCOMO II model to HPC setups. We lay out a general methodology to evaluate HPC projects with COCOMO II and analyse its cost parameters for the investigated parallelisation projects with OpenACC on NVIDIA GPUs. Further, we evaluate the accuracy of the model in comparison to the reported efforts of the projects and investigate the impact of inaccuracies in the cost parameter ratings by means of a global sensitivity analysis.

Keywords: COCOMO; OpenACC; GPU; development effort; effort estimation; sensitivity analysis.

DOI: 10.1504/IJCSE.2018.095849

International Journal of Computational Science and Engineering, 2018 Vol.17 No.3, pp.283 - 296

Received: 18 Jan 2017
Accepted: 23 Apr 2017
Published online: 25 Oct 2018 *

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