Comparative review of cloud computing platforms for data science workflows

Abstract

The proposed framework for the comparative review of cloud computing platforms for data science workflows uses an amalgamation of the analytical hierarchy process, Saaty’s fundamental scale of absolute numbers, and a selection of relevant evaluation criteria (namely: automation, error handling, fault tolerance, performance quality, unit testing, data encryption, monitoring, role based access, security, availability, ease of use, integration and interoperability) to enable users to evaluate criteria pertaining to cloud platforms for data science workflows. Three cloud providers were chosen as alternatives for selection by the user to deter mine which cloud platform would be most suitable for the user. The analytical hierarchy process used two levels, namely criteria and sub-criteria to ensure that attributes selected as sub-criteria were relevant and had a material impact on alternative selection. Experiments con ducted and subsequent evaluation of criteria revealed that out of the three anonymised cloud service providers (CSP1, CSP2 and CSP3), CSP3 was most suitable for the user based on their valuation of the criteria. Out of the three alternate providers, CSP3 had an importance weighting of 0.53, CSP1 had 0.28 and CSP2 had 0.19, indicating that CSP3 would be most suitable for the evaluator in terms of the criteria considered. Thus, the proposed framework is successfully able to recommend which cloud platform would be suitable for the user based on the relative importance of the above criteria. Evaluations of the criteria are shown to be consis tent and thus the weighting of criteria against the goal of cloud platform selection are sensible. The proposed framework is robust enough to accommodate for changes in criteria and alternatives, depending on user cloud platform requirements and scope of cloud platform selection.