Cluster Analysis for Characterising Computer System Workloads - Panacea or Pandora?