Individual Data Plan in Virtual Network Operation: A Proactive Matching Approach

Abstract

High-quality cellular wireless connection has become a necessity for today’s internet users, who however, have to pay high prices when they use conventional ISPs’ networks. To alleviate this pain, a new type of “mobile virtual network operators which purchases data plans from multiple ISPs and resells them to individuals in an on-demand way, has emerged. A great challenge for such virtual network operators is to determine how to “match” available data plans to dynamically changing users’ demands in an intelligent and instant manner. To this end, we propose CAPITAL, a proactive resource allocation framework based on behavior-driven individual cellular data usage prediction. First, based on measurement studies of a real-world virtual network provider over 2 years covering 3 million users, we reveal the dynamical cellular data demands of users, and the spatial, temporal and behavioral factors that affect individuals’ data usage; Second, we propose a context-aware prediction model based on wide & deep network, to capture individual data usage; Third, we formulate the data plan allocation as an assignment problem, and use a shortest augmenting path algorithm to solve it efficiently. Our trace-driven experiments show that our design can reduce the cellular data cost by 78.6% for the network operator, while retaining the same user experience level.