Workload models and performance evaluation of cloud storage services

Cloud storage systems are currently very popular with many companies offering services, including worldwide providers such as Dropbox, Microsoft and Google. These companies as well as providers entering the market could greatly benefit from a deep understanding of typical workload patterns their services have to face in order to develop cost-effective solutions. Yet, despite recent studies of usage and performance of these systems, the underlying processes that generate workload for the system have not been deeply studied. This paper presents a thorough investigation of the workload generated by Dropbox customers. We propose a hierarchical model that captures user sessions, file system modifications and content sharing patterns. We parameterize our model using passive measurements gathered from fourdifferent networks. Next, we use the proposed model to drive the development of CloudGen, a new synthetic workload generator that allows the simulation of the network traffic created by cloud storage services in various realistic scenarios. We validate CloudGen by comparing synthetic traces with actual data from operational networks. We then show its applicability by investigating the impact of the continuing growth in cloud storage popularity on bandwidth consumption. Our results indicate that a hypothetical 4-fold increase in both user population and content sharing could lead to 30 times more network traffic. CloudGen is a valuable tool for administrators and developers interested in engineering and deploying cloud storage services