Measurement and Analysis of a Streaming Media Workload

The increasing availability of continuous-media data is provoking a significant change in Internet workloads. For example, video from news, sports, and entertainment sites, and audio from Internet broadcast radio, telephony, and peer-to-peer networks, are becoming commonplace. Compared with traditional Web workloads, multimedia objects can require significantly more storage and transmission bandwidth. As a result, performance optimizations such as streaming-media proxy caches and multicast delivery are attractive for minimizing the impact of streaming-media workloads on the Internet. However, because few studies of streaming-media workloads exist, the extent to which such mechanisms will improve performance is unclear. This paper (1) presents and analyzes a client-based streaming-media workload generated by a large organization, (2) compares media workload characteristics to traditional Web-object workloads, and (3) explores the effectiveness of performance optimizations on streaming-media workloads. To perform the study, we collected traces of streaming-media sessions initiated by clients from a large university to servers in the Internet. In the week-long trace used for this paper, we monitored and analyzed RTSP sessions from 4,786 clients accessing 23,738 distinct streaming-media objects from 866 servers. Our analysis of this trace provides a detailed characterization of streaming-media for this workload.