Towards Intelligent Edge Storage Management: Determining and Predicting Mobile File Popularity

The exponential growth in smart device usage and mobile applications markets, coupled with the increased complexity of these applications has outpaced the storage typically available on these devices. As a result, developers and end-users have both started pursuing cloud solutions for the storage and caching of application data, as well as personal user data. Such solutions, however, require a great deal of manual intervention from the user to first determine which files need to be offloaded, and then perform the actual task of offloading them. Given the time, attention, and effort required for such manual tasks, there is a need for automated storage management systems that make these decisions in an intelligent manner. In this paper, we analyze the access traces of 28 different mobile device users, and show that there are discernible access patterns to both popular and unpopular files on a user's device. We leverage this knowledge of access patterns to develop our Pattern Based Popularity Assessor (PBPA) algorithm. Our evaluation shows that PBPA can predict file popularity with high accuracy, making it a promising candidate for future automated mobile storage offloading systems.