Abstract
The amount of traffic in wireless networks is increasing exponentially, and this problem can be mitigated using device-to-device (D2D) caching technology, which installs a cache on a mobile end device. Devices can reduce the cell load through self-offloading via content in their own cache and D2D offloading using content in others' caches. However, especially in the early stage of D2D caching systems, a limited number of devices with a small storage might be used, and it is required to develop a caching scheme with excellent performance despite the small cache size. Regarding content popularity, which is common to most users, the preference probability values are not concentrated on some pieces of content, making it difficult to achieve satisfactory performance using a small cache. On the other hand, when considering individual users, content preferences may contain large values for specific content based on individual characteristics. In addition, the performance can be improved by considering short-term content preferences that reflect changes in content preferences over time or newly created content during peak hours. In this article, the hit ratio is divided into six parts considering self- and D2D offloading, common and individual user preferences, and little and large temporal changes in content preferences during peak hours. We also conceptually divide the cache of a helper into six areas in relation to the six parts of the hit ratio, and discuss cache partitioning and proactive caching strategies according to the environment.
Highlights
Due to the increase in content-based services such as video streaming, the amount of traffic in wireless networks is exponentially growing [1]–[4]
In this article, considering common and personal, and static and dynamic characteristics, content preferences are divided into four parts, namely, static common, static personal, dynamic common and dynamic personal preferences, and we discuss how they should be considered
The four parts of content preferences of device n at time 0, PSC (k), PSP(n, k), PDC (k, 0), and PDP(n, k, 0) are all Zipf distributions with Zipf parameter of 0.8, but the numbers of non-zero elements, KSNPonzero(n), KDNConzero, and KDNPonzero(n) are limited to 500, 1000, and 20, respectively, while KSNConzero is 100,000. This means that dynamic and/or personal preferences are assumed to have larger values on some specific content, while static common preference values are spread over a large number of pieces of content
Summary
Due to the increase in content-based services such as video streaming, the amount of traffic in wireless networks is exponentially growing [1]–[4]. It covers the content storage and delivery, cell load, device mobility, and content preference models used in this article, in addition to discussing the hit ratio and cache partitioning. This article assumes that a helper can provide data to any device in the same cell using multi-hop D2D communication or utilizing a base station as a relay. If the mobile helper does not go through the cell to which device n belongs, we let Min(n) = Mout (n) = 0 In this case, we assume that the following equation holds:. Let Snomadic denote the set of the nomadic devices in the cells other than the target cell, and Smobile denote the set of the mobile devices other than the mobile helper In this case, the following equations are assumed to hold: Mocuetll. Using the device mobility model described in this subsection, we discuss the caching strategies of nomadic or mobile helpers
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
