Let Once-Request Data Go: An Online Learning Approach for ICN Caching

Abstract

In-network caching significantly improves the efficiency of data transmission in ICN by replicating requested data for future re-access. In this work, we shift our focus on once-request data, which cannot be re-used and would lead to under-utilization of in-network caching. We present a name feature-based online learning approach to recognizing and filtering once-request data when making caching decision. It can dynamically update its parameters through online observation on previous recognition. Evaluation results show that our learning approach can recognize once-request data with more than 80% accuracy. By filtering those data, 76% cache replacement operations are saved and cache hit ratio is increased by 151%.