160 likes | 297 Vues
Understanding Online Social Network Usage from a Network Perspective. F. Schneider et al (T-Labs, AT&T) Internet Measurement Conference 2009 Networking Journal Club 26th Feb 2010. Outline. Introduction OSN features Data set Methodology Feature Popularity OSN Session Characteristics
E N D
Understanding Online Social Network Usagefrom a Network Perspective F. Schneider et al (T-Labs, AT&T) Internet Measurement Conference 2009 Networking Journal Club 26th Feb 2010
Outline • Introduction • OSN features • Data set • Methodology • Feature Popularity • OSN Session Characteristics • Dynamic within OSN Sessions
Introduction • OSN: Half a billion users • Facebook adds over 377,000 users every day • Interesting: • For ISPs (transport data and provide connectivity) • For OSN service providers (develop and operate scalable systems) • For researchers and developers (indentify trends and improve designs)
Introduction • Objectives: • Which features of OSN are popular and capture the users attention? • What is the impact of OSNs on the network? • What needs to be considered whe designing future OSNs? • Is the user’s behavior homogeneous?
OSN features • Profiles, friends, photos, videos, messaging, groups… third-party applications • Example:
OSN features • Active (user’s clicks) vs. Indirect (auto triggered, e.g. ajax, javascripts, images…) • Using of other locations/servers
Data set • 6 http header traces • 2 ISPs • 20,000 DSL users: 2500(6000) use OSN. • DAG cards • 4 OSNs: Facebook, StudiVZ, Hi5, LinkedIn
Methodology • Extract OSN session clickstream • To do this, they need to be able to know OSN traffic: manual traces • With manual traces: • Site names • Cookies • HTTPS (or not) • Handshakes • Signatures and patterns
Lessons Learned (manual traces) • Reverse-engineering user interactions with OSNs from HTTP traces is non-trivial • Major bottleneck: manual traces for validation and classification of the rr-pairs • Number of patterns for each OSN is large (253, 218, 206 and 299). • OSN restructures => new patterns • HTTPS • Tamper Data plug-in for Firefox very useful
Feature popularity • Which OSN features are more fascinating? Does it depend on OSN? • ISP needs to care as it might impact bandwidth demand • OSN needs to care as it might impact server resources
Bytes per session/Duration • Heavy-tailed distribution: Weibull • Session size between 200KB and 10MB • High Variability: mean of about 40 min • More than 10% last longer than 1 hour • Peak between 5 sec and 2 min
Summary • Reconstruct OSN clickstreams from HTTP header traces from passive monitoring. • Methodology for identifying OSN sessions and user actions within the OSN. • Indentify the features that are important to the users. • Generate workloads for evaluating novel OSNs.