Journal of Information & Systems Management Yue Yong, Olaniyi Olawale Omoyajowo 16 3 2026 https://doi.org/10.6025/jism/2026/16/3/110-125 https://www.dline.info/jism/fulltext/v16n3/jismv16n3_2.pdf Effective workload characterization is essential for resource management in cloud data centres, yet the relative importance of different workload descriptors remains unclear. This study analyses a corpus of 15,000 workload observations to identify the dominant empirical dimensions of workload variation and to evaluate whether secondary behavioural metrics provide independent discriminatory information beyond read/write orientation. Observations were classified into read heavy (74.77%), mixed (25.09%), and writeheavy (0.14%) categories. We further cross classified a six type workload taxonomy with four data modalities to examine finer descriptive structure. Principal Component Analysis followed by silhouette optimised Kmeans clustering (K = 2, silhouette = 0.1957) partitioned the feature space into two groups distinguished almost exclusively by read/write composition, while request rate, hotspot intensity, access skew, data locality, and workload change rate remained virtually identical across clusters. Kruskal Wallis testing confirmed that read and write ratios differed highly significantly among workload classes (H = 8493.179, p < .001), whereas all secondary metrics were statistically homogeneous (p > .20). The findings demonstrate that read/write orientation constitutes the principal organising axis of workload behaviour in this corpus, while secondary characteristics do not form independent classification dimensions. The results support a parsimonious, hierarchical workload representation in which read/write ratio serves as the primary classification feature and secondary metrics function as continuous operational descriptors. This paper discusses implications for workload aware database selection, cloud native architecture, and resource management frameworks.