240 likes | 360 Views
Modeling peer-peer file sharing systems. Ge, Z.; Figueiredo, D.R.; Sharad Jaiswal; Kurose, J.; Towsley, D.; INFOCOM 2003. Twenty-Second Annual Joint Conference of the IEEE Computer and Communications Societies. Presentation by Tony Sung, MC Lab, IE CUHK 23th September 2004. Outline.
E N D
Modeling peer-peer file sharing systems Ge, Z.; Figueiredo, D.R.; Sharad Jaiswal; Kurose, J.; Towsley, D.;INFOCOM 2003. Twenty-Second Annual Joint Conference of the IEEE Computer and Communications Societies. Presentation by Tony Sung, MC Lab, IE CUHK23th September2004
Outline • Significance of the Paper • Modeling P2P Systems • Capturing the Parameters • The Results • Conclusion
Significance of the Paper Mathematical Model that addresses • Scalability • Impact of freeloaders • Impact of imbalanced service capacity and file request load
Significance of the Paper Results • Limited-scope flooding worsen performance • Freeloaders do not have significant impact • Centralized indexing approach can scale gracefully up to its capacity • Degradation due to replication-popularity mismatch is minor
Modeling P2P Systems Modeling
Modeling P2P Systems 3 types of systems • CIA – Centralized Indexing Architecture • DIFA – Distributed Indexing with Flooding Architecture • DIHA – Distributed Indexing with Hashing Architecture
Modeling P2P Systems • A multiple class closed queuing network • C classes of peers: N = (N (1),…, N (C)) • Mnumber of distinct files sorted according to availability • States • Thinking, Querying,Downloading, Offline
Capturing the Parameters Capturing Parameters
Capturing the Parameters A single class case, C = 1, Na= (Na(1)) • (Popularity, Zipf, assumed …) • (File service rate) • ? (Common services’ service rate) • ? (Query Failure Rate)
Capturing the Parameters Common services’ service rate • CIA: • DIFA: • DIHA:
Capturing the Parameters Query failure rate • CIA: • DIFA: • DIHA:
Capturing the Parameters Freeloaders: 2 classes
Solving the model Visiting Ratio Service Demand Bottleneck Queue
Results Results
Results Scalability DIHA DIHA CIA CIA DIFA DIFA
Results Freeloader’s impact on system throughput Non-freeloaderpopulation
Results Freeloader’s impact on non-freeloader throughput Non-freeloaderpopulation
Results Reason of DIFA’s poor performance
Results Mismatch in file popularity and replication
Results Mismatch in file popularity and replication
Conclusion • Simple, representative math. Model • Flexible across architectures • Dealt with different classes of peers
Results Freeloader’s impact on system throughput - another perspective