Geo-Edge: Geographical Resource Allocation on Edge Caches for Video-on-Demand Streaming

Geographical information has shown great potential in optimizing the resource allocation for Video-on-Demand (VoD) systems, e.g., the VoD service provider can allocate more bandwidth/computing resources to certain regions where more user requests are generated. Recently, the deployment of edge caches close to client users has become a cost-effective solution to the large-scale VoD system, as popular video content can be placed in edge caches so as to reduce the response latency of user requests, and save the bandwidth consumption on CDN edge servers. In this paper, we propose an edge-cache-assisted VoD system, named Geo-edge. The system develops a machine learning approach—a modified LSTM (Long Short Term Memory) network-to predict the amount and distribution of requests to each video, and proactively place appropriate video resources in the buffer of edge caches. We conduct emulations over the traces of real-world video sessions in ten different autonomous systems in the backbone network of China. Results show that the length of the path between source and clients could decrease by 29% with the help of edge caches, while Geo-Edge can significantly save the expenditure of CDN bandwidth under two typical charging policies of Internet service providers (i.e., a reduction of 45% when charged by overall throughput, or 35% when charged by peak bandwidth consumption).