Category-based user interaction with online user-generated videos: workload characterization

Caching video content in content distribution networks takes advantage of repeated requests amongst a population of users. The efficiency of replication and cache placement policies depends on which user requests which video object. We develop a methodology for analyzing user interactions, based on video category. We then analyze a trace file of YouTube requests captured at a university campus from the client side of the network. Though the dataset has suffered from content deletion, enough information is available to give some insight into users' viewing patterns. Differentiation in the relative amount of repeated viewing suggests that category-specific distribution and caching policies may provide more efficient use of computational resources. Initial analysis suggests that user access patterns can be identified to a better extent when video category is considered.