Longitudinal Analysis of Major Video Streaming Services in the US
Donna Hooshmand, Reza Rejaie PhD
Department of Computer and Information Sciences,University of Oregon,
1.Introduction 3.Methods 4.Project Status
v Video streaming applications (e.g., v First, raw data must be turned into the proper format AKA v This is an ongoing project that was 
Netflix, Hulu,…) have become cooked data. Here are some main features of a selected Daily started 1 month ago. Currently, a 
increasingly popular over the Internet in snapshot of a UOnet netFlow data showing the time, size, parser is being developed and is 
the past decade. autonomous system (AS), and internet protocol (IP). estimated to be fully operational soon.
v It is important to gain insight into their v For now, the working parser parses the 
relative popularity and other data in per seconds, per minute, and 
characteristics. per hour intervals. 
v Ascertaining traits of popular v Currently a few methods of finding the 
applications is beneficial to best way to map IPs to ASes are being 
determining what aspects tested.
contribute to an applications v From the support for the summer 
popularity. v Running a parser on this data can isolate and filter the through (NSF-REU), the research will 
v This study relies on the data from UOnet information that is relevant to our question. be focusing on fully developing the 
to perform a longitudinal analysis on the v The created parser consists of different modules that parser and doing a statistical analysis 
characteristics of popular internet can be specified by the user to get relevant data to determine what makes applications 
applications. needed to answer the proposed questions. popular. 
2.Research Question v For example, if one module is for finding the organization for a References
flow, then this module maps the IP to it’s AS, which can then be v A View From the Edge: A Stub-AS 
In this study we will answer some mapped to the organization. Perspective of Traffic Localization and its 
questions including the following: v Previous studies at UO that have used this technique. Implications 
vWhat percentage of UO traffic is Here is the result of one focused on traffic localization: Bahador Yeganeh, Reza Rejaie,    
comprised of video streaming v This figure depicts Walter Willinger
applications (e.g., Netflix, Hulu)? the volume of IEEE/IFIP Network Traffic Measurement   
vWhat is the duration of the time users delivered traffic and Analysis Conference (TMA), Dublin,  
use that application? from individual Ireland, June 2017
v How has the bandwidth (or popularity) content providers [acceptance rate 35%]
of video streaming applications changes (CPs) to UOnet One of the top three papers selected for
along with the the best paper award.over the past few years?
CDF of aggregate v (The provided graphs are from this research).vUsing machine learning and clustering 
algorithms, could we determine fraction of traffic by Acknowledgments
whether a connection is associated with top 21 CPs in the I’d like to thank Reza Rejaie, PhD students 
a specific video streaming application snapshot  from 10/04/16. in ONRG, Namely Soheil Jamshidi, and 
based on its network level signature? v Using these same methods, it is possible to determine how Christopher Misa for their guidance and 
various features affect popularity. assistance throughout this study.