
Research Article
Analyzing Aggregate User Behavior on a Large Multi-platform Content Distribution Service
@INPROCEEDINGS{10.1007/978-3-030-98005-4_12, author={Raushan Raj and Adita Kulkarni and Anand Seetharam and Arti Ramesh and Antonio A. de A. Rocha}, title={Analyzing Aggregate User Behavior on a Large Multi-platform Content Distribution Service}, proceedings={Ad Hoc Networks and Tools for IT. 13th EAI International Conference, ADHOCNETS 2021, Virtual Event, December 6--7, 2021, and 16th EAI International Conference, TRIDENTCOM 2021, Virtual Event, November 24, 2021, Proceedings}, proceedings_a={ADHOCNETS \& TRIDENTCOM}, year={2022}, month={3}, keywords={}, doi={10.1007/978-3-030-98005-4_12} }
- Raushan Raj
Adita Kulkarni
Anand Seetharam
Arti Ramesh
Antonio A. de A. Rocha
Year: 2022
Analyzing Aggregate User Behavior on a Large Multi-platform Content Distribution Service
ADHOCNETS & TRIDENTCOM
Springer
DOI: 10.1007/978-3-030-98005-4_12
Abstract
In recent years, Video on Demand (VoD) streaming has increased exponentially as a result of reduced streaming costs and higher bandwidth. For retention of consumers, it is crucial for content providers to understand the behavior of their users and continuously improve performance. In this paper, we analyze the user behavior onGlobo.com, the largest content distribution service in Brazil. We consider 1.4 billion logs spanning a period of four weeks from October 25, 2020 to November 21, 2020. We analyze the user request patterns and the trends in server’s response time. We explore metrics such as protocol, status code, cache hits, user agent, content category popularity and geographical distribution of users. We finally investigate the video popularity distribution and trends in size of content downloaded. We observe that the highest number of requests occur between 8 pm and 11 pm. We observe that 57% of requests are served over HTTPS, while significant portion (43%) are still served over HTTP. Our analysis also reveals that nearly 97% of requests result in a cache hit. Additionally, we observe that the video popularity distribution is skewed and follows a power law with 10% of the videos accounting for 87% of the requests.