Docstoc

BitTorrent An Extensible Heterogeneous Model

Document Sample
BitTorrent An Extensible Heterogeneous Model Powered By Docstoc
					            BitTorrent:
An Extensible Heterogeneous Model

 Alix L.H. Chow University of Southern California
 Leana Golubchik University of Southern California
 Vishal Misra Columbia University
                    Infocom 2009
             Presented by Zhanfeng Wang
                                           Contents
•     Introduction
•     BT Model
•     Model validation and insight
•     Applications
•     Conlusion



PLA University of Science and Technology
Network Engineering Laboratory
                                              Introduction
                               Free-rider




            Seed                                  transfer
                                                  chunks     Leecher


                                           peer




PLA University of Science and Technology
Network Engineering Laboratory
                               Introduction(cont.)
• BT’s Two phase
         – Initial phrase (make-span or flash crowd)
         – Steady state
• Methodology
         – Markov model
• Limitation
         – homogeneous


PLA University of Science and Technology
Network Engineering Laboratory
                               Introduction(cont.)
• Ignored by the other literatures
         – Seeds come from:
                  • Unmonitored BT clients
                  • communities enforce a download/upload ratio
         – Seeding behavior effect
                  • Compensate for the asymmetric bandwidth
                    scenarios in the Internet
                  • degrade the fairness and incentive properties of
                    the system

PLA University of Science and Technology
Network Engineering Laboratory
                               Introduction(cont.)
         – Free-riders’ opportunity
                  • capacity provided by leechers through the
                    optimistic unchoking mechanism
                  • capacity provided by seeds




PLA University of Science and Technology
Network Engineering Laboratory
                               Introduction(cont.)
• Ignored by the other literatures
         – BT clients are prone to “find” like nodes that
           are similar in bandwidth capacity and
           exchange chunks amongst each other.




PLA University of Science and Technology
Network Engineering Laboratory
                               Introduction(cont.)
• Contribution
         – our model is distinct as it accounts for: (1)
           seeds and(2) free-riders.
         – improve the accuracy of our model,we
           account for: (a) imperfect clustering
           behavior unchokes and (b) bias in optimistic
           unchokes.



PLA University of Science and Technology
Network Engineering Laboratory
                                           Contents
•     Introduction
•     BT Model
•     MODEL VALIDATION AND INSIGHT
•     Applications
•     Conlusion



PLA University of Science and Technology
Network Engineering Laboratory
                                           BT Model
• Terminology
         – H: be the number of node classes, each class is
           defined by its upload and download capacities,Ui and
           Di, respectively.
         – λ:the average rate of the new nodes arrive to the
           system
         – pi: probability of that a newly arrived node belongs
           to class i, thus, class i nodes arrive at rate of λi = piλ.
         – Nil: the leechers number of class i
         – Nis: the seed number of class i
         – Ni = Nil + Nis represents the total number of class i
           nodes

PLA University of Science and Technology
Network Engineering Laboratory
                                           BT Model
• Terminology
         – Til: the average amount of time a class i
           leecher takes to download a file
         – Tis:average amount of time a class i node
           stays in the system after becoming a seed.
         – Ti = Til + Tis: the average amount of time a
           class i node stays in the system
         – di =m/Ti:

PLA University of Science and Technology
Network Engineering Laboratory
                                           BT Model
• Primitive model




Hypotheses:
1. the system is in steady state
2. Perfect clustering among nodes of the same class
3. The optimistic unchokes are unbiased

PLA University of Science and Technology
Network Engineering Laboratory
                                           BT Model
• Primitive model
         – Regular Unchokes:



         – Optimistic Unchokes:



         – Seed Unchokes:

PLA University of Science and Technology
Network Engineering Laboratory
                                           BT Model
• Primitive model :




PLA University of Science and Technology
Network Engineering Laboratory
                                           BT Model
• Free-Riders’ download speed




PLA University of Science and Technology
Network Engineering Laboratory
                                           BT Model
• More Realistic Model
         – take into account the fact that a fraction of
           regular unchokes will go to nodes in other
           classes.




PLA University of Science and Technology
Network Engineering Laboratory
                                           BT Model
• More Realistic Model
         – consider the fact that optimistic unchokes of
           a node are not distributed evenly to all
           leechers.




PLA University of Science and Technology
Network Engineering Laboratory
                                           BT Model
• More Realistic Model




                            How to determin qi,j and oi,j?


PLA University of Science and Technology
Network Engineering Laboratory
                                           BT Model
• Imperfect Clustering




PLA University of Science and Technology
Network Engineering Laboratory
                                            ReDiR
                                           • the average number of optimistic unchokes
                                           that a slow node is receiving from fast nodes.
• Imperfect Clustering




PLA University of Science and Technology
Network Engineering Laboratory
                                           BT Model
• Biased Optimistic Unchoking




PLA University of Science and Technology
Network Engineering Laboratory
                                           Contents
•     Introduction
•     BT Model
•     Model validation and insight
•     Applications
•     Conlusion



PLA University of Science and Technology
Network Engineering Laboratory
     MODEL VALIDATION AND INSIGHT
   Two Leecher Classes




PLA University of Science and Technology
Network Engineering Laboratory
     MODEL VALIDATION AND INSIGHT
 Three Leecher Classes




PLA University of Science and Technology
Network Engineering Laboratory
     MODEL VALIDATION AND INSIGHT
 Free Riders




PLA University of Science and Technology
Network Engineering Laboratory
     MODEL VALIDATION AND INSIGHT
      Peer set size




PLA University of Science and Technology
Network Engineering Laboratory
                                           Contents
•     Introduction
•     BT Model
•     Model validation and insight
•     Applications
•     Conlusion



PLA University of Science and Technology
Network Engineering Laboratory
                                           Applications
• Large view exploit




PLA University of Science and Technology
Network Engineering Laboratory
                                           Applications
• Large view exploit




PLA University of Science and Technology
Network Engineering Laboratory
                                           Applications
• Distributing seeding capacity
• Sort-based (N):
   – Unchokes the N which are furthest from the
     middle (based on sorting
• Threshold-based (K,N):
   – N chosen are from those which have a
     certain percentage of the total number of
     chunks


PLA University of Science and Technology
Network Engineering Laboratory
                                           Applications
• Distributing seeding capacity
• Threshold Optimization:
   – ma = mK and mb = m(1 − K).




PLA University of Science and Technology
Network Engineering Laboratory
                                           Applications
• Distributing seeding capacity




PLA University of Science and Technology
Network Engineering Laboratory
                                           Applications
• Distributing seeding capacity




PLA University of Science and Technology
Network Engineering Laboratory
                                           Applications
• Distributing seeding capacity




PLA University of Science and Technology
Network Engineering Laboratory
                                           Applications
• Distributing seeding capacity




PLA University of Science and Technology
Network Engineering Laboratory
                                           Contents
•     Introduction
•     BT Model
•     Model validation and insight
•     Applications
•     Conlusion



PLA University of Science and Technology
Network Engineering Laboratory
                                           Conclusion
• The model can be used by the community
  to gain insights on the working of BT and
  help design improvements.




PLA University of Science and Technology
Network Engineering Laboratory
                                           BT Model
• The model can be used by the community
  to gain insights on the working of BT and
  help design improvements.




PLA University of Science and Technology
Network Engineering Laboratory

				
DOCUMENT INFO
Shared By:
Categories:
Tags:
Stats:
views:13
posted:7/11/2011
language:English
pages:38