Characterizing User Association Patterns for Optimizing Small-Cell Edge System Performance

Edge computing is a promising paradigm to support multifarious time-sensitive applications. In this article, we shed light on data-driven approaches to optimize edge system performance via mining user association patterns in the wireless local area network (WLAN). Particularly, we first describe the association traces (containing more than 50,000 users) that are collected in one operational WLAN network. We then conduct the data analytics to mine user association patterns that have impacts on edge system performance. To leverage our findings, we propose three data-driven approaches to optimize the edge system performance, that is, efficient edge resource deployment, mobility-aware user service migration, and distributed cooperative learning for edge intelligence. Finally, we cast a case study on distributed learning by devising a cooperation scheme, named co-location time scheme (CoLo) (e.g., C-Located learning), which can leverage the user association patterns to distribute learning tasks. Extensive data-driven experiments corroborate the efficacy of CoLo in comparison with state-of-the-art schemes.

Automated summary

This article explores data-driven approaches to optimize edge system performance by mining user association patterns in WLAN. The study describes the collected association traces and analyzes the impact of user association patterns on edge system performance. Three data-driven approaches are proposed, including efficient resource deployment, mobility-aware user service migration, and distributed cooperative learning for edge intelligence. A case study on distributed learning validates the effectiveness of the proposed cooperation scheme, CoLo.