AI

Optimizing the update packet stream for web applications

Abstract

The Internet has evolved to an extent where users now expect any-where any-time and any-form access to their personalized data and applications of choice. However providing a coherent (seamless) user experience across multiple devices has been relatively hard to achieve. While the 'how to sync' problem has been well studied in literature, the complementary 'when to sync' problem has remained relatively unexplored. While frequent updates providing higher user satisfaction/retention are naturally more desirable than sparse updates, the steadily escalating resource costs are a significant bottleneck. We thus propose extensions to the traditional periodic refresh model based on an adaptive 'smart sync approach' that enables variable rate updates closely modeling expected user behavior over time. An experimental evaluation of the proposed mechanism on a sizeable subset of users of the GMAIL web interface indicates that the proposed refresh policy can achieve the best of both worlds - limited resource provisioning and minimal user-perceived delays.