Not All Clouds Are Transparent: Handling Unavailable Attributes in CSP Selection

Abstract

Selecting a cloud service provider (CSP) is a complex task for enterprises, as each CSP offers a distinct set of quality of service (QoS) attributes with varying values [1]. These attributes include durability, response time, best practices, throughput, availability, compliance, latency, reliability, and successability [2]. Traditionally, researchers have employed a multi-attribute decision-making (MADM) algorithm to select a suitable CSP, operating under the assumption that complete QoS attribute values are available. However, certain QoS attribute values may be unavailable in many CSPs. This lack of values makes the selection process non-transparent for enter prises, hindering their ability to identify the most suitable CSP. Consequently, the unavailable values can be imputed using either simple or advanced techniques to facilitate the selection of a suitable CSP. Common imputation techniques include mean, minimum, maximum, regression, k-nearest neighbour, and rough set theory. After the imputation process, an MADM algorithm can then be applied to identify the most suitable CSP. Such MADM algorithms include the technique for order preference by similarity to ideal solution (TOPSIS), the best holistic adaptable ranking of attributes technique (BHARAT), and multi-objective optimization on the basis of ratio analysis