Where academic tradition
meets the exciting future

You are here: TUCS > PUBLICATIONS > Publication Search > Analysis and Transcoding Time ...

Analysis and Transcoding Time Prediction of Online Videos

Tewodros Deneke, Sebastien Lafond, Johan Lilius, Analysis and Transcoding Time Prediction of Online Videos. TUCS Technical Reports 1145, TUCS, 2015.

Abstract:

With the current development of user generated media content, sharing video files
and video streams is an essential and growing part of business in the content and
media industry. Today, video content is delivered to a myriad of devices over dif-
ferent communication networks. Video delivery must be adapted to the available
bandwidth, the screen size, resolution and the decoding capability of the end de-
vices. As a result, very large-scale video processing and transcoding services and
applications are deployed to serve a growing number of video delivering requests
for numerous type of devices. With relatively large variations in the required com-
putational effort for changing the format, resolution and bit rate of a video, large
scale video transcoding platforms require efficient load balancing techniques in
order to maximize the system utilization rate. This become essential when such
platforms are deployed in the cloud.
In this work we present a novel approach for the challenging task of predicting
the transcoding time of a video into another given transcoding parameters and an
input video. The prediction of the video transcoding time can be used to make
effective load balancing decisions on a large scale video transcoding platforms
serving several thousands of transcoding requests at a time. To obtain enough
information on the characteristics of real world online videos and their transcod-
ing parameters needed to model transcoding time, we built a video characteristics
dataset, using data collected from a large video-on-demand system, YouTube. The
dataset contains a million randomly sampled video instances listing 10 fundamen-
tal video characteristics. The data was collected based on an unbiased sampling
method, the random prefix sampling [1]. Our analysis on the dataset provides
insightful statistics on fundamental video characteristics and can be further ex-
ploited to optimize or model components of a multimedia processing systems.
We further present models, based on support vector machines, linear regres-
sion and multi-layer perceptron feedforward artificial neural network, for predict-
ing transcoding time of videos based on the selected input video features and
transcoding parameters. Extensive experimental evaluation and analysis on var-
ious test videos and transcoding parameters demonstrate the effectiveness of the
proposed approach.

BibTeX entry:

@TECHREPORT{tDeLaLi15a,
  title = {Analysis and Transcoding Time Prediction of Online Videos},
  author = {Deneke, Tewodros and Lafond, Sebastien and Lilius, Johan},
  number = {1145},
  series = {TUCS Technical Reports},
  publisher = {TUCS},
  year = {2015},
  keywords = {Video Characteristics Dataset, Benchmark, Transcoding, Prediction},
  ISBN = {978-952-12-3297-8},
}

Belongs to TUCS Research Unit(s): Embedded Systems Laboratory (ESLAB)

Edit publication