Delivery Strategies for Dynamic Contents in Edge Caching Systems

Abstract

In today's digital age, multimedia consumption is pervasive, driven by video-on-demand platforms, online social networks, and sensor networks. Whether it is the latest blockbuster movie on a streaming service or real-time data from environmental sensors, one commonality remains: content is in a constant state of evolution. This evolution is encapsulated under the umbrella term of dynamic content, denoting the inherent variability and temporal nature of content.

This thesis focuses on optimal caching and delivery of dynamic contents. In the second and third chapters of the thesis, we study caching of dynamic contents that arrive at the server and sojourn for random times and have time-varying popularity while they live. In the fourth chapter, content is updated over time, and its values are attributed to its freshness.

We begin by studying content caching in a cellular network consisting of a Base Station with a cache. New contents arrive in the network according to a Poisson process and the contents stay for exponentially distributed times. The contents’ request rates are time-varying, their instantaneous request rates being a decreasing function of the number of contents in the network. At any given time, all the contents in the network have the same popularity or request rate. Precaching contents at the Base Station incurs a cost but fetching contents from the server on being requested incurs an even higher cost. We formulate the caching problem as a Markov decision process and derive the optimal caching policy. We also propose a Reinforcement Learning based algorithm that yields precaching decisions when the system parameters are unknown. We show, via numerical illustrations, that the proposed algorithm's time-averaged cost is close to the optimal average cost.

Next, we study content caching in a scenario where the contents stochastically arrive in the system and stay for random times. While a content is alive, its popularity evolves according to a continuous time Markov chain. Different contents can have different popularities and hence the request rates. As in Chapter 2, fetching contents from the server and caching at the Base Station cache incur a cost. However, not having the requested contents at the Base Station incurs an even higher cost. We study optimal proactive caching to minimise the time-average content miss and caching costs. We solve the single content caching problem using the framework of average cost Markov decision problems and formulate the multi-content problems as a restless multi-armed bandit problem with mortal bandits. We show that the problem is indexable and explicitly derive the Whittle indices. We demonstrate the performance of the Whittle index policy via numerical evaluation.

Finally, we study content update and delivery in an edge caching system consisting of a sensor, a set of users, and an aggregator. The users have an aversion to stale contents, quantified as the contents' version age, and they occasionally request the aggregator for updated contents. The aggregator fetches the contents from the sensor and serves the users on demand. We study optimal content fetching and serving problem, aiming to minimize the time-averaged content fetching, transmission and age costs. This problem also lends itself to the Markov decision problem framework. We derive the optimal policy in the special case of single user. We use this solution to offer a heuristic for the general multi-user problem which has a high dimensional state space due to different content ages at different users. We demonstrate via simulations that the heuristic performs well.