Transitioning from on-premise computing to cloud computing A cost comparison case study on a Swedish grocery retail company

Abstract: Using public cloud platforms for managing applications and compute is becoming increasingly common among organizations. In this thesis, a case study was performed at a Swedish grocery retailer that was in pursue of migrating all analytics related IT-resources hosted on both on-premise and private cloud to a public cloud platform. The main purpose of the thesis was to investigate if a public cloud would be more cost efficient than a private cloud and the main research question was: "Is it cost efficient for Swedish grocery retail companies to utilize public clouds compared to utilizing private clouds in terms of storage and computational costs?". The two platforms compared were a public cloud platform (referred to as PCP) and a private on-premise cloud (referred to as POPC).  One of the main selling arguments for transitioning from on-premise to a public cloud is cost savings. However, this argument has not been well investigated, and the applicability of this argument across various contexts remains unclear. Therefore, this thesis objective was to investigate this cost savings argument where the cost of storage and compute resources were the main focus. The thesis also aimed to question the notion that "public beats private" as a universal maxim.  To answer the research question, both platforms were compared from a cost perspective by performing an experiment. A set of SQL-queries run as a pipeline was run in similar environments on both platforms. Since Spark was used as the computing technology on the private cloud, a Spark cluster was chosen as the computing environment on both platforms. Furthermore, the case study also included data collection in terms of semi-structured and unstructured interviews in addition to analysis of internal documents about the two platforms concerning costs and investigations concerning the public cloud transition decision.   The results from the experiment indicated that the selected Spark application was approximately 20% cheaper to run on PCP. In addition, findings from interviews and documents indicate that the main reasons for transitioning to a public cloud was not driven by cost savings. Instead, strategic arguments such as positioning for future requirements, opportunity costs, faster adoption to new technologies and scaleable resources were presented as the most important arguments. Nevertheless, cost savings was still an important argument for making the transition decision. Therefore, in conclusion, running the selected Spark application on PCP was cheaper than running the same application on the POPC platform's Spark cluster. Hence, Spark clusters on PCP can be considered cost competitive compared to Spark clusters on POPC platforms in terms of storage and compute. However, to reach a general conclusion whether public clouds are cost efficient compared to private clouds in terms of storage and compute would require further investigation. In addiiton, it was also concluded that evaluating only the cost perspective of public cloud platforms versus on-premise platforms is inadequate when making a cloud transition decision.