LSTM Network Assisted Content Caching at the Edge for Video on Demand
Mahmoud Darwich, Magdy Bayoumi · 2023
Video-on-demand (VoD) services often suffer from long start-up delays and poor streaming quality due to distance from centralized data centers. In this work, we develop a long short-term memory (LSTM) network to predict future video requests and intelligently cache content at the edge for optimized VoD delivery. We train the LSTM model on a large proprietary dataset of VoD access logs capturing real-world sequential patterns. Using caching simulations, our approach increases cache hit ratios by 56% compared to recency-based caching and 41% over popularity-based schemes. By proactively caching predicted videos at the edge, our method reduces average video startup time by 32% and decreases upstream bandwidth consumption by 48%. These substantial gains demonstrate the value of forecasting VoD popularity using LSTMs and harnessing edge computing to cache content closer to viewers. Our intelligent edge caching system can improve quality of experience for users and lower costs for providers.