Abstract With the rapid evolution of information technology, global network traffic has grown exponentially, challenging the efficacy of traditional centralized storage for big data management. Consequently, distributed storage systems (DSSs) have emerged as a crucial solution, while also enabling real-time intelligent decision-making at the edge. The content delivery network (CDN) is a prominent instance of such DSSs. In this paper, we investigate the storage resource allocation problem for DSS within the CDN context, explicitly incorporating system reliability into the model. We propose a profit-oriented model tailored to a cloud-edge collaborative CDN framework that holistically integrates content revenue, storage cost, and repair cost from the perspective of network operators. To optimize storage allocation for contents with heterogeneous sizes and popularity (hotness), we develop an efficient solution by combining greedy and backtracking algorithms. Furthermore, we integrate pruning techniques within the backtracking algorithm to drastically curb computational complexity. Comparative experiments against traditional strategies, including multi-copy storage and uniform redundancy schemes, demonstrate the consistent profitability superiority of our approach. The performance advantages are especially pronounced in scenarios with high system reliability requirements, elevated edge storage costs, and substantial revenue from high-speed content delivery. This work contributes to enabling reliable data storage for distributed edge intelligence.
Gao et al. (Wed,) studied this question.