Outline Simulating overlay networks with PeerSim ! Introduction to Peersim – What is Peersim – Peersim components ! Aggregation – What is aggregation – Average in Peersim Andrea Marcozzi 06/03/2008 Dipartimento di Scienze dell’Informazione, Università di Bologna ! SLAC – Tags – What is SLAC – SLAC in Peersim 0 1 What is PeerSim? Introduction: P2P Systems ! ! Peer-to-Peer systems are potentially huge (composed of millions of nodes); ! Nodes join and leave the network continuously; ! Evaluating a new protocol in a real environment is not an easy task ! ! ! ! ! ! 2 PeerSim is an open source P2P systems simulator developed at the Department of Computer Science, University of Bologna It has been developed with Java Available on Source Forge (peersim.sf.net) Its aim is to cope with P2P systems properties High Scalability (up to 1 million nodes) Highly configurable Architecture based on pluggable components 3 Network Representation (I) The Peersim Simulation Engine ! Peersim Node 1 has two simulation modes: – Event-driven: a set of events (messages) are scheduled at different simulation time and the node protocols are run accordingly to messages’ delivery times – Cycle-driven: sequential simulation, in each cycle each protocol’s actions are executed by every node (we will focus on this) 2 3 ! Network: global array which contains all the network nodes ! Node: each node’s state and actions are described through a stack of protocols. Protocols are accessed through a Pid ! Linkable: interface used to access and manage node’s view n-2 n-1 n Network Node Protocol 1 (Application) ! Protocol 0 (Linkable) ! 4 Network Representation (II) ……… A node is a stack of protocols node.getProtocol(pid) 5 Network Representation (III) ! CDProtocol: interface used to describe node’s actions at each cycle. A generic node can both perform local actions (CDProtocol) and manage the local view (Linkable) ! Control: performs the global initialization and performances analysis 6 7 Main Interfaces: Protocol ! The CDProtocol interface is used to define cycle-driven protocols, that is the actions performed by each node at each simulation cycle ! Each node can run more than one protocol ! Protocols are executed sequentially Main Interfaces: Linkable ! Linkable is used to manage node’s view. Typical actions are: – Add neighbor – Get neighbor – Node’s degree – Remove neighbor 8 9 Peersim tree Main Interfaces: Control ! SRC ! Control is used to define operations that require global network knowledge as: – Initialization of the network and protocols – Network dynamism – Statistical analysis (observer) 10 contains the source code of Peersim and of some example protocols (aggregation, newscast) ! The parameters for the simulation are specified by a Configuration File 11 Example: AverageFunction 1. AverageFunction 0. Newscast Aggregation By Aggregation we mean calculating a certain function (eg. average) over a number of values distributed over a network 12 Example: AverageFunction (II) ! Each node has a numeric value ! The aggregate value has to be known by each node ! Local value of node i contains current estimate of the average ! Node i select random peer j: – i and j exchange estimates – i and j update local estimate with the average 14 13 Configuration File 01 #random.seed 1150540268549 02 simulation.cycles 300 03 control.shf Shuffle 04 network.size 50000 05 06 protocol.0 example.newscast.SimpleNewscast 07 protocol.0.cache 20 08 protocol.1 example.aggregation.AverageFunction 09 protocol.1.linkable 0 10 order.protocol 0 1 11 12 init.0 peersim.dynamics.WireKOut 13 init.0.protocol 0 14 init.0.k 20 15 16 init.1 example.aggregation.PeakDistributionInitializer 17 init.1.protocol 1 18 init.1.value 10000 19 20 control.ob0 aggregation.AverageObserver 21 control.ob0.protocol 1 General settings Protocols settings Control settings 15 Config File: Protocol Settings Config File: General Settings 06 protocol.0 example.newscast.SimpleNewscast 07 protocol.0.cache 20 08 protocol.1 example.aggregation.AverageFunction 09 protocol.1.linkable 0 10 order.protocol 0 1 01 #random.seed 1150540268549 02 simulation.cycles 300 03 control.shf Shuffle 04 network.size 50000 ! ! ! ! Line 01: define the seed to be used. If not defined uses a random seed; Line 02: defines the number of cycles Line 03: if defined, shuffles the Network array at the beginning of each cycle Line 04: defines the Network size ! ! ! ! ! Line 06: define protocol 0 as Newscast Line 07: define protocol 0’s variable cache (it is the view size) Line 08: define protocol 1 as the application calculating average aggregation Line 09: indicates that the linkable is Newscast (0) Line 10: indicated the order for protocols execution 16 Config File: Control Settings 17 Aggregation: execution 12 init.0 peersim.dynamics.WireKOut 13 init.0.protocol 0 14 init.0.k 20 15 16 init.1 example.aggregation.PeakDistributionInitializer 17 init.1.protocol 1 18 init.1.value 10000 19 20 control.ob0 aggregation.AverageObserver 21 control.ob0.protocol 1 ! ! ! Lines 12 - 14: protocol 0 initialization Lines 16 - 18: protocol 1 initialization Lines 20 - 21: protocol 1 observer 18 On the ReadMe file in the Peersim home directory there are the instructions for running the example 19 Peersim Scheduler ! Hence a Peersim simulation is scheduled in this way: Initializers execution while (time < cycles) for each node in the Network CDProtocol actions are executed end foreach Control actions are executed end while 20 Implementing AverageFunction 01. public void nextCycle(Node node, int protocolID) { 02. int linkableID = FastConfig.getLinkable(protocolID); 03. Linkable linkable = (Linkable) node.getProtocol(linkableID); 04. if (linkable.degree() > 0) { 05. Node peer = linkable.getNeighbor(CommonState.r .nextInt(linkable.degree())); 06. 07. // Failure handling 08. if (!peer.isUp()) return; 09. 10. AverageFunction neighbor = (AverageFunction)peer.getProtocol(protocolID); 11. double mean = (this.value + neighbor.value) / 2; 12. this.value = mean; 13. neighbor.value = mean; } } 21 SLAC: Introduction ! SLAC is an evolutionary algorithm whose aim is to promote Cooperation in P2P systems ! Inspired by Computational Sociology (Tag systems) ! Tested with the “Prisoner’s Dilemma” SLAC (Selfish Link and behavioural Adaptation to produce Cooperation) – Simple 2 player game – Represents contradiction between selfishness of the players and global benefit 22 23 What are Tags? The Prisoner’s Dilemma (PD) Player 1 1.9 1 0.1 0 Strategy DC CC DD CD ! Studied by computational sociology (Holland 1992) ! Tags are observable markings (hair style, clothes…) ! Tags evolve just like any other artificial trait ! Limiting interactions between agents with similar tags leads to cooperative altruistic behaviour Player 2 0 1 0.1 1.9 !D = Defect; C = Cooperate ! PD Constraints: – DC > CC > DD > CD – 2CC > DC+CD 24 25 Tag Systems’ agents ! ! ! Each agent in a Tag system is identified by a “Tag” (it may be a set of bits) Each agent has a “behaviour” (it may simply be to cooperate or defect) Each agent has an “utility” Mutation ! Mutation of the tag: – Changing some traits (i.e. some bits) ! Mutation of the strategy: – Flipping the strategy bit (from cooperation to defection or vice-versa) - Selfish optimization: agents copy tags and behaviour of those performing better (with higher utility) - Periodically tags and strategies are mutated with a certain probability 26 27 Tag systems’ evolutionary algorithm From Tags to Networks (P2P) ! Agents are nodes ! Tags are represented by the node’s view (we consider this as a kind of tag) LOOP some number of generations LOOP for each agent (a) in the population Select a game partner agent (b) Agent (a) and (b) invoke their strategies receiving the appropriate payoff END LOOP Reproduce agents in proportion to their average payoff Apply mutation to tag and strategy of each reproduced agent with low probability END LOOP Interaction phase - Evolution phase – Interaction is between neighbors ! Behaviour – File sharing (share files or free riding) – Prisoner’s Dilemma (cooperate or defect) ! Utility ! ! 28 SLAC is an attempt to move tags in P2P networks Nodes performs some application task (file sharing, PD) and get some utility The nodes move to find better neighbors 29 SLAC Pseudocode Each node i periodically with a certain probability: j := selectRandomPeer(); if U j >U i drops all its current links links to node j and copy j’s strategy and links mutate (with low probability) links and strategy end if – The links and the strategy of the nodes with better performances (higher utility) are copied – The topology of the network evolves ! evaluated at application level – File sharing: number of files downloaded – PD: outcome given by the payoffs matrix SLAC ! defined by the application Groop-like selection between clusters of nodes – Groups of good guys spread – Selfish nodes become isolated 30 ! Mutation: i drops all its links and connects to a random node Random Peer given by Newscast 31 Copy and Rewire - Mutation SLAC simulations ! SLAC has been implemented in Peersim on top of Newscast (random sampling for utility comparison) ! The abstract application used for testing SLAC is the PD – Each node has a strategy (C / D) – At each Peersim cycle nodes play a single round with a random neighbor ! At the end of each cycle, SLAC is invoked 32 33 SLACER (I) SLACER (II) ! SLACER is a variant of SLAC ! In SLAC when a node moves, it first drops all its link ! In SLACER when a node moves, it drops its links with a certain probability W ! If W = 1, then SLACER == SLAC 34 ! Drop probability value leads to different kind of topology: – W = 1, highly partitioned network (disconnected); – W = 0.9, SmallWorld-like network – Low W, random network (more robust, less cooperation) 35 SLAC(ER) architecture #random.seed 1150540268549 simulation.cycles 300 control.shf Shuffle network.size 4000 Three protocol: Newscast, SLACER, PDapp 2. PDapp (Application) Cooperative neighbor Random node (movement) Configuration File (I) utility protocol.0 example.newscast.SimpleNewscast protocol.0.cache 20 protocol.1 slacer.Slacer protocol.1.degree 20 protocol.1.linkable 0 protocol.1.application 2 protocol.1.reprprob 0.20 protocol.1.mutation 0.001 protocol.1.tagmutation 0.01 1. SLACER 0. Newscast (Peer Sampling Service) General settings Protocols settings protocol.2 slacer.PdApp.PdProtocol protocol.2.linkable 1 36 37 Configuration File (II) SLAC experiments (I) init.0 peersim.dynamics.WireKOut init.0.protocol 0 init.0.k 20 init.1 slacer.SlacerInitializer init.1.protocol 1 init.1.isolateprob 1 init.1.copyprob 1 init.2 slacer.PdApp.PdStrategyInitializer init.2.protocol 2 init.2.quantity 0 Control settings init.3 peersim.dynamics.WireKOut init.3.protocol 1 init.3.k 20 control.ob0 slacer.BooleanStrategyObserver control.ob0.protocol 2 control.ob1 slacer.PdApp.PdObserver control.ob1.protocol 2 38 Cycles=4000; ReprRate=0.20; Single run 39 SLAC experiments (II) Internet resources ! Peersim resources page: http://peersim.sf.net ! These slides on: http://www.cs.unibo.it/~marcozzi ! Other similar slides: http://www.cs.unibo.it/~arteconi ! For any questions: [email protected] Cycles for high cooperation (93%) 40 References Hales, D. and Arteconi, S. (2006) Slacer: A self-organizing protocol for coordination in p2p networks. IEEE Intelligent Systems ! Jelasity, M., Montresor, A. and Babaoglu, O. (2005) Gossip-based aggregation in large dynamic networks. In ACM Transaction on Computer Systems (TOCS). Volume 23, Issue 3 ! Marcozzi, A.; Hales, D.; Jesi, G.; Arteconi, S.; Babaoglu, O. (2005) Tag-Based Cooperation in Peer-to-Peer Networks with Newscast. In Self- Organization and Autonomic Informatics (I), Volume 135 Frontiers in Artificial Intelligence and Applications, Edited by: H. Czap, R. Unland, C. Branki and H. Tianfield, IOS Press, Netherlands ! - Some of them available as UBLCS technical report at: www.cs.unibo.it/research/reports/42 41
© Copyright 2024