Peer-to-Peer Networks 08 Kelips and Epidemic Algorithms

Slides:



Advertisements
Similar presentations
Fast Convergence of Selfish Re-Routing Eyal Even-Dar, Tel-Aviv University Yishay Mansour, Tel-Aviv University.
Advertisements

RUMOR SPREADING “Randomized Rumor Spreading”, Karp et al. Presented by Michael Kuperstein Advanced Topics in Distributed Algorithms.
Gossip Algorithms and Implementing a Cluster/Grid Information service MsSys Course Amar Lior and Barak Amnon.
Feb 7, 2001CSCI {4,6}900: Ubiquitous Computing1 Announcements.
Rumors and Routes Rajmohan Rajaraman Northeastern University, Boston May 2012 Chennai Network Optimization WorkshopRumors and Routes1.
Gossip algorithms : “infect forever” dynamics Low-level objectives: – One-to-all: Disseminate rumor from source node to all nodes of network – All-to-all:
CPSC 689: Discrete Algorithms for Mobile and Wireless Systems Spring 2009 Prof. Jennifer Welch.
Rumour Spreading in Social Networks Alessandro Panconesi Dipartimento di Informatica Joint work with Flavio Chierichetti and Silvio Lattanzi.
©2002; Aberer,Datta,Hauswirth; LSIR-IC-EPFL. Laboratoire de systèmes d'informations répartis Updates: in a highly unreliable environment. Motivation: Maintaining.
Fast Distributed Algorithm for Convergecast in Ad Hoc Geometric Radio Networks Alex Kesselman, Darek Kowalski MPI Informatik.
CPSC 689: Discrete Algorithms for Mobile and Wireless Systems Spring 2009 Prof. Jennifer Welch.
ETH Zurich – Distributed Computing Group Roger Wattenhofer 1ETH Zurich – Distributed Computing – Christoph Lenzen Roger Wattenhofer Exponential.
1 University of Freiburg Computer Networks and Telematics Prof. Christian Schindelhauer Mobile Ad Hoc Networks Network Coding and Xors in the Air 7th Week.
Karger’s Min-Cut Algorithm Amihood Amir Bar-Ilan University, 2009.
1 University of Freiburg Computer Networks and Telematics Prof. Christian Schindelhauer Wireless Sensor Networks 19th Lecture Christian Schindelhauer.
1 University of Freiburg Computer Networks and Telematics Prof. Christian Schindelhauer Distributed Coloring in Õ(  log n) Bit Rounds COST 293 GRAAL and.
1 University of Freiburg Computer Networks and Telematics Prof. Christian Schindelhauer Mobile Ad Hoc Networks Theory of Data Flow and Random Placement.
1 University of Freiburg Computer Networks and Telematics Prof. Christian Schindelhauer Mobile Ad Hoc Networks Mobility (III) 12th Week
An Approximation Algorithm for Requirement cut on graphs Viswanath Nagarajan Joint work with R. Ravi.
On the Crossing Spanning Tree Vineet Goyal Joint work with Vittorio Bilo, R. Ravi and Mohit Singh.
Dept. of Computer Science Distributed Computing Group Asymptotically Optimal Mobile Ad-Hoc Routing Fabian Kuhn Roger Wattenhofer Aaron Zollinger.
Maximal Independent Set Distributed Algorithms for Multi-Agent Networks Instructor: K. Sinan YILDIRIM.
Advanced Topics in Algorithms and Data Structures 1 Two parallel list ranking algorithms An O (log n ) time and O ( n log n ) work list ranking algorithm.
CPSC 689: Discrete Algorithms for Mobile and Wireless Systems Spring 2009 Prof. Jennifer Welch.
Multicast Communication Multicast is the delivery of a message to a group of receivers simultaneously in a single transmission from the source – The source.
Decomposing Networks and Polya Urns with the Power of Choice Joint work with Christos Amanatidis, Richard Karp, Christos Papadimitriou, Martha Sideri Presented.
Communication (II) Chapter 4
Proof of Kleinberg’s small-world theorems
Distributed Coloring Discrete Mathematics and Algorithms Seminar Melih Onus November
Small clique detection and approximate Nash equilibria Danny Vilenchik UCLA Joint work with Lorenz Minder.
2.1 Computational Tractability. 2 Computational Tractability Charles Babbage (1864) As soon as an Analytic Engine exists, it will necessarily guide the.
1 HEINZ NIXDORF INSTITUT University of Paderborn Algorithms und Complexity Seminar Designing Peer-to-Peer- Networks Christian Schindelhauer
HEINZ NIXDORF INSTITUTE University of Paderborn Algorithms and Complexity Algorithms for Radio Networks Exercise 11 Stefan Rührup
1 Permutation routing in n-cube. 2 n-cube 1-cube2-cube3-cube 4-cube.
HEINZ NIXDORF INSTITUTE University of Paderborn Algorithms and Complexity Online Routing in Faulty Meshes with Sub-linear Comparative Time and Traffic.
PODC Distributed Computation of the Mode Fabian Kuhn Thomas Locher ETH Zurich, Switzerland Stefan Schmid TU Munich, Germany TexPoint fonts used in.
1 HEINZ NIXDORF INSTITUTE University of Paderborn Algorithms and Complexity Christian Schindelhauer Algorithms for Radio Networks Winter Term 2005/2006.
An Efficient Decentralized Algorithm for the Distributed Trigger Counting (DTC) Problem Venkatesan T. Chakravarthy (IBM Research-India) Anamitra Roy Choudhury.
Vertex Coloring Distributed Algorithms for Multi-Agent Networks
Peer-to-Peer Networks 03 CAN (Content Addressable Network) Christian Schindelhauer Technical Faculty Computer-Networks and Telematics University of Freiburg.
CSCE 668 DISTRIBUTED ALGORITHMS AND SYSTEMS Spring 2014 Prof. Jennifer Welch CSCE 668 Set 3: Leader Election in Rings 1.
Complexity and Efficient Algorithms Group / Department of Computer Science Testing the Cluster Structure of Graphs Christian Sohler joint work with Artur.
Andrea CLEMENTI Radio Networks The Model Broadcast.
1 Distributed Vertex Coloring. 2 Vertex Coloring: each vertex is assigned a color.
1 HEINZ NIXDORF INSTITUTE University of Paderborn Algorithms and Complexity Christian Schindelhauer Algorithms for Radio Networks Winter Term 2005/2006.
Peer-to-Peer Networks 08 Kelips and Epidemic Algorithms Christian Schindelhauer Technical Faculty Computer-Networks and Telematics University of Freiburg.
Robust Random Number Generation for Peer-to-Peer Systems Baruch Awerbuch Johns Hopkins University Christian Scheideler Technical University of Munich.
Peer-to-Peer Networks 05 Pastry Christian Schindelhauer Technical Faculty Computer-Networks and Telematics University of Freiburg.
Randomized Algorithms for Distributed Agreement Problems Peter Robinson.
Peer-to-Peer Networks 04: Chord Christian Schindelhauer Technical Faculty Computer-Networks and Telematics University of Freiburg.
Small-world phenomenon: An Algorithmic Perspective Jon Kleinberg.
Peer-to-Peer Networks 09 Random Graphs for Peer-to-Peer-Networks Christian Schindelhauer Technical Faculty Computer-Networks and Telematics University.
Theory of Computational Complexity Probability and Computing Chapter Hikaru Inada Iwama and Ito lab M1.
Peer-to-Peer Networks 07 Degree Optimal Networks
Probabilistic Algorithms
Randomized Min-Cut Algorithm
A Log-Star Distributed Maximal Independent Set Algorithm for Growth-Bounded Graphs Johannes Schneider Roger Wattenhofer TexPoint fonts used in EMF.
New Characterizations in Turnstile Streams with Applications
Monitoring Churn in Wireless Networks
Wireless Sensor Networks 5. Routing
Minimum Spanning Tree 8/7/2018 4:26 AM
Peer-to-Peer Networks 09 Random Graphs for Peer-to-Peer-Networks
Gossip-based Data Dissemination
Peer-to-Peer Networks 07 Degree Optimal Networks
Know thy Neighbor’s Neighbor Better Routing for Skip Graphs and Small Worlds Moni Naor Udi Wieder.
Peer-to-Peer and Social Networks
MST in Log-Star Rounds of Congested Clique
Turnstile Streaming Algorithms Might as Well Be Linear Sketches
TexPoint fonts used in EMF.
Proof of Kleinberg’s small-world theorems
Presentation transcript:

Peer-to-Peer Networks 08 Kelips and Epidemic Algorithms Christian Schindelhauer Technical Faculty Computer-Networks and Telematics University of Freiburg

Random Call Model In each round a new contact graph Gt=(V,Et): Each node in Gt has out-degree 1 chooses random node v out of V Infection models: Push-Model if u is infected and (u,v) ∈ Et, then v is infected in the next round Pull-Modell: if v is infected and (u,v) ∈ Et, then u is infected in the next round 2

Push Model 3

Push Model 4

Push Model 5

Push Model 6

Push Model 7

Push Model 8

Push Model 9

Push Model 10

Push Model Start Phase 3 cases for an infected node it is the only one infecting a new node it contacts an already infected node it infects together with other infected nodes a new node this case is neglected in the prior deterministic case Probability for 1st or 3rd case s(t) = 1-i(t) Probability for 2nd case i(t) Probability for 3rd case is at most i(t) since at most i(t) are infected Probability of infection of a new node, if i(t) ≤ s(t)/2: at least 1 – 2i(t) E[i(t+1)] ≥ i(t) + i(t)(1 – 2 i(t)) = 2i(t) -2i(t)2 ≈ 2 i(t) 14

Push Model Start phase & Exponential Growth   15

Push Model Middle Phase & Saturation Probability of infections of a new node if i(t) ≤ s(t)/2: 1 – 2i(t) E[i(t+1)] ≥ 2 i(t) – 2i(t)2 ≈ 2 i(t) Middle phase I(t) ∈ [n/(log n), n/3] term 2i(t)2 ≥ 2i(t)/(log n) cannot be neglected anymore Yet, 2i(t) – 2i(t)2 ≥ 4/3 i(t) still implies expontential growth, but with base < 2 Saturation: I(t) ≥ n/3 Probability that a susceptible node is not contacted by I(t) = c n infected nodes: This implies a constant probability for infection ≥ 1 – e–1/3 und ≤ 1 – e– 1 Hence E[s(t+1)] ≤ e–i(t) s(t) ≤ e–1/3 s(t) Chernoff-bounds imply that this holds with high probability Exponential shrinking of susceptible nodes Base converges to 1/e 19

Push Model 20

Random Call Model Infection models: Push Model Pull Model if u is infected and (u,v) ∈ Et, then v is infected in the next round Pull Model if v is infected and (u,v) ∈ Et, then u is infected in the next round 21

Pull Model 22

Pull Model 23

Pull Model 24

Pull Model 25

Pull Model 26

Pull Model 27

Pull Model Consider an susceptible node and I(t) infected nodes Probability that a susceptible node contacts an infected node: i(t) E[s(t+1)] = s(t) – s(t) i(t) = s(t) (1 – i(t)) = s(t)2 E[i(t+1)] = 1-s(t)2 = 1 – (1 – i(t))2 = 2 i(t) – i(t)2 ≈ 2 i(t) for small i(t) 30

Pull Model Problem if i(t) ≤ (log n)2 then exponential growth is not with high probability O(log n) steps are needed to start eh growth with high probability yet in the expectation it grows exponentially After this phase If s(t) ≤ ½ then the share of susceptible nodes is squared in each step This implies E[s(t+ O(log log n))] = 0, If i(t) ≥ ½ then after O(log log n) steps all nodes are infected with high probability 31

Pull Model i(t) 32

Push&Pull Model Combines growth of Push and Pull Start phase: i(t) ≤ 2 c (ln n)2 Push causes doubling of i(t) after O(1) rounds with high probability Exponential growth: I(t) ∈ [2 c (ln n)2, n/(log n)] Push and Pull nearly triple in each round with high probability: i(t+1) ≥ 3 (1-1/(log n)) i(t) Middle phase: I(t) ∈ [n/(log n), n/3] Push and Pull slower exponential growth Quadratic shrinking: I(t) ≥ n/3 caused by Pull: E[s(t+1)] ≤ s(t)2 The Chernoff bound implies with high probability s(t+1) ≤ 2 s(t)2 so after two rounds for s(t) ≤ 1/21/2 s(t+2) ≤ s(t)2 w.h.p. 33

Push&Pull Model 34

Max-Counter Algorithm Simple termination strategy If the rumor is older than maxctr, then stop transmission Advantages simple Disadvantage Choice of maxctr is critical If maxctr is too small then not all nodes are informed If maxctr is too large, then the message overhead is Ω(n maxctr) Optimal choice for push-communication maxctr = O(log n) Number of messages: O(n log n) Pull communication Push&Pull communication maxctr = log3n + O(log log n) Number of messages: O(n log log n) 35

Shenker‘s Min-Counter Algorithm Only is the rumor is seen as old then contact partners increase the age- counter Shenkers Min-Counter-Algorithmus für maxctr = O( log log n) Every player P stores age-variable ctrR(P) for each rumor R A: player P does not know the rumor: ctrR(P) ← 1 B: If player P sees rumor for the first time ctrR(P) ← 1 B: If partners Q1, Q2, …, Qm  communicate with P in a round If mini{ctrR(Qi)} ≥ ctrR(P) then ctrR(P) ← ctrR(P) + 1 C: If ctrR(P) ≥ maxctr then tell the rumor for maxctr more rounds then D: stop sending the rumor Theorem Shenkers Min-Counter algorithms informs all nodes using Push&Pull-communication in log3n + O(log log n) rounds with probability 1−n−c, using at most O(n log log n) messages. 36

Shenker‘s Min-Counter-Algorithm Theorem Shenkers Min-Counter algorithms informs all nodes using Push&Pull- communication in log3n + O(log log n) rounds with probability 1−n−c, using at most O(n log log n) messages. 37

Peer-to-Peer Networks 08 Kelips and Epidemic Algorithms Christian Schindelhauer Technical Faculty Computer-Networks and Telematics University of Freiburg