Interest Forwarding Strategy in Named Data Networks using Thompson Sampling

Abstract

Optimizing Interest forwarding and Data delivery has been among the top dissected problems in NDN for the last decade; however, only a few contributions thrive to minimize communication cost and delay concurrently. In NDN, a receiver-driven forwarding strategy is considered resource-consuming as the routers incur computation to find the best path to the desired item, specified by an Interest’s name. On the other hand, a source-driven forwarding strategy, a scheme that suppresses the sub-optimal sources, experiences increased delay when no source answers in the exploration phase. The confluence of the two strategies can counteract the drawbacks of each one, which, however, has never been investigated. In this work, a reinforcement learningbased, namely Thompson Sampling, strategy is proposed that operates in a receiver-cum source driven fashion to optimize Interest forwarding and answering. The proposed method introduces a ’Beam’ concept coupled with adaptive scoped-flooding to optimize Interest forwarding, and the sources adopt Thompson Sampling to suppress the sub-optimal responses. When hit by an Interest, an optimal source sends back the desired Data to the consumer whereas a sub-optimal source remains Silent. Together, the ’Beam’ and the scoped-flooding adapt the Interest forwarding range based on cache hit/miss ratio. The adaptation optimizes communication cost and delay, and contributes to scheming the proposed strategy resource-savvy. The proofof- concept implementation in software (simulation) reveals that the proposed system outperforms the counterpart benchmarks by reducing the communication costs and delay in NDN (by around 350% and 10%, respectively) without negotiating packet delivery ratio.