On High Performance Cloud Based File Synchronization with User Collaboration

Abstract

Over the past few years, cloud-based file storage/synchronization systems like Dropbox, Gdrive and Skydrive, have achieved tremendous success among internet users. This new generation of service, beyond conventional client/server or peer-to-peer file hosting with storage only, provides reliable file storage and effective file synchronization for diverse user collaborations. In this thesis, we take a close look to understand such cloud-based file synchronization and collaboration systems. Using Dropbox as a case study, our real-world measurement carefully decomposes its file synchronization protocol into different stages: {\it pre-processing}, {\it uploading}, {\it downloading}, and {\it post-processing}. We show that these series of computation and communication operations, which is far more complicated than those in conventional file hosting, is necessary for Dropbox-like services especially considering the cloud deployment. Such a design can significantly improve service reliability and avoid the possible task interference on cloud-based virtual machines (VMs). Unfortunately, these operations also lead to higher latency and cost. In particular, the variance of latency across different users increases with larger population, and thus individual users may face severe performance degradation when the system scale grows. Moreover, we also notice that Dropbox assumes that their users are not online at the same time. The files are therefore uploaded to a cloud storage server and then pushed to the destination. It is easy to see that such a design is inefficient when some of the Dropbox users are online at the same time. To address this problem, we propose an enhancement to let Dropbox detect user's online status and decide whether we can directly send them the file. We tested our prototype on {\it PlanetLab} and the evaluation indicates that the design can greatly reduces the file synchronization latency with minimal system overhead.