O uso de histórico de acesso na definição de taxa de bits em sessões de vídeo com taxa adaptável

Abstract

Video related traffic is prevalent on the Internet. By 2022, this traffic will be responsible for 82% of all data transfer throughout the Internet. The HTTP Adaptive Streaming (HAS) is the most popular technology behind modern video applications. Its millstone approaching is to use data collected by clients to build sessions that balance image quality and continuity. The connection between HAS-based clients and servers is a crucial factor in finding that balance. A number of these clients use data from ongoing sessions. In these scenarios, continuous measurements and a clear understanding of essential elements are crucial to estimate the connection throughput due to its natural instability. However, client-side events have the potential to steam slots of silence in those measurements that drive the embedded algorithms for adaptive streaming to erratic behavior. In this work, the concept of achievable rate is introduced based on data collected by proactive monitors. Machine Learning methods are applied to those data to build throughput predictive models, and numerical studies are carried to assess its accuracy. The assessed accuracy shows that the conceived models can improve the decision made by those algorithms that use ongoing sessions data to find the balance between session continuity and image quality.