Characterizing web-based video sharing workloads

Abstract

Video sharing services that allow ordinary Web users to upload video clips of their choice and watch video clips uploaded by others have recently become very popular. This paper identifies invariants in video sharing workloads, through comparison of the workload characteristics of four popular video sharing services. Our traces contain meta-data on approximately 1.8 million videos which together have been viewed approximately 6 billion times. Using these traces, we study the similarities and differences in use of several Web 2.0 features such as ratings, comments, favorites, and propensity of uploading content. In general, we find that active contribution, such as video uploading and rating of videos, is much less prevalent than passive use. While uploaders in general are skewed with respect to the number of videos they upload, the fraction of multi-time uploaders is found to differ by a factor of two between two of the sites. The distributions of life-time measures of video popularity are found to have heavy-tailed forms that are similar across the four sites. Finally, we consider implications for system design of the identified invariants. To gain further insight into caching in video sharing systems, and the relevance to caching of life-time popularity measures, we gathered an additional data set tracking views to a set of approximately 1.3 million videos from one of the services, over a twelve week period. We find that life-time popularity measures have some relevance for large cache (hot set) sizes (i.e., a hot set defined according to one of these measures is indeed relatively "hot"), but that this relevance substantially decreases as cache size decreases, owing to churn in video popularity.

Characterizing and modelling popularity of user-generated videos

“…Studies have examined the characteristics of user-generated video files [6,7,11,16,21], use of social networking features in video-sharing services [13,16], the structure of YouTube's "friend" network [15], the use of the "video response" feature of YouTube [4], the popularity characteristics of user-generated videos [5,6,10,11,16,21], and also models for user-generated video popularity prediction [14,19]. Here, we restrict attention mostly to related work on popularity characterization and modelling for user-generated videos.…”

“…Prior work on user-generated video popularity characterization has typically relied either on network traffic traces from a network gateway [11,21] or meta-data sampled from video sharing services [5,6,10,16]. Both Gill et al [11,12] and Zink et al [21] analyzed YouTube video requests from a campus network and observed that the video requests follow a Zipf-like distribution and network bandwidth savings may be feasible if large proxy caches are used.…”

Dynamic file bundling for large-scale content distribution

“…As in our steady-state experiments, we use two scenarios: one with purely random file requests, in which the popular file is requested with higher probability than the less popular ones, and one in which the files requested by each node are statically assigned. The results for both cases are similar, and for the purpose of evaluation, only results for (1) actual (2) max (3)(4)(5)(6)(7)(8)(9)(10) mean (3)(4)(5)(6)(7)(8)(9)(10) min (3)(4)(5)(6)(7)(8)(9)(10) Fig. 7: Number of active downloaders of each file type, using the random bundling policy for the dynamic experiments.…”

“…Characterization studies of file popularity [1]- [5] have shown that there typically is a long tail of mildly and less popular content. Companies like Amazon (books) and Apple (music) have leveraged this long tail of niche content to increase sales of more popular items as well.…”

End-to-End Resource Management for Federated Delivery of Multimedia Services

“…As time passes, it will be able to more accurately estimate values for these parameters, allowing it to iteratively improve the solution using the subsequent steps of the algorithm. The modelling and prediction of popularity distribution curves of multimedia services has been an active research topic for several years [38,39]. Existing methods can thus be applied to estimate the cumulative popularity distribution.…”