A Survey of Current YouTube Video Characteristics

raffic produced by YouTube has had a significant impact on both fixed and mobile networks. The study and evaluation of YouTube content features can benefit network traffic engineering by supporting the development of sustainable video delivery services and regulation of network traffic. Such evaluations are particularly useful to network operators who aim to refine and optimize existing cache algorithms to better adapt to YouTube video traffic patterns. The main objective of this article is to depict the latest YouTube traffic profiles and deliver updated and valuable information for future researchers. There has been a healthy amount of research on YouTube video analysis, but most of it was conducted prior to 2007. Following Google’s acquisition of YouTube in 2008, several major aspects of the network and service framework were restructured, leading to changes in the user policy and service infrastructure—including new limits on video duration, size, and resolution. Here, we reflect on YouTube video characteristics in light of such changes. Given that it has been more than six years since similar studies were conducted, our findings provide comparative insight into how YouTube videos have developed over recent years. Measurement Methodology We compare statistics from the works of Xu Cheng, Cameron Dale, and Jiangchuan Liu; Phillipa Gill and her colleagues; and Abdolreza Abhari and Mojgan Soraya with our own findings, because these works present similar scopes of investigation, covering video categories, video durations, file sizes, and bit rates. All four publications were based on statistics from prior to Google’s acquisition of YouTube, so we use a comparative analysis to depict how YouTube videos have developed in recent years based on the collective statistics (see Table 1). In terms of YouTube traffic analysis, a variety of research has been performed with different approaches and objectives. Pablo Ameigeiras and his colleagues collected statistics from the most viewed video clips, studied the impact of video encoding rates on traffic generation, and proposed a traffic model that could be used for feasible and effective simulation. However, we don’t use this work as a reference point, because it was based on a total trace of 32,860 videos, and it analyzed traffic from a different perspective.