Peer-to-Peer Algorithms and Prototypes in Jyv

1
Peer-to-Peer Algorithms and Prototypes in
Jyväskylä
Presentation for Workshop on Peer-to-Peer
Networking10.10.2005

• Mikko Vapa, research studentmikvapa_at_jyu.fiDepart
  ment of Mathematical Information Technology
• University of Jyväskylä, Finland
• http//tisu.it.jyu.fi/cheesefactory

2
Contents

• Peer-to-Peer Algorithms
• Formalization of Peer-to-Peer Resource Discovery
  Problem
• Approximation of Optimum for P2P Resource
  Discovery Algorithms using k-Steiner Minimum
  Trees
• NeuroSearch P2P Resource Discovery Using
  Evolutionary Neural Networks
• Peer-to-Peer Prototypes
• Chedar Peer-to-Peer Middleware
• Mobile Chedar
• Peer-to-Peer Studio
• Peer-to-Peer Distributed Computing
• Mobile Peer-to-Peer Encounter Networks
• Gasoline Price Comparison System and BlueCheese
• Development History and Future

3
Peer-to-Peer Algorithms
4
Formalization of P2PResource DiscoveryProblem

• Currently, only textual definitions of P2P
  resource discovery problem exist given a
  resource name, find the node or nodes that manage
  the resource
• Textual definitions are poor, because they do not
  precisely tell
• What kind of a graph is used for finding
  resources
• What information is locally available to nodes
  taking part in the finding process
• Therefore, the task of forwarding a resource
  query is unclear

5
Formalization of P2P Resource Discovery Problem

• Formalization can be used for
• Formalization of peer-to-peer resource discovery
  algorithms
• Breadth-First Search
• Highest Degree Search
• Evaluating the performance of peer-to-peer
  resource discovery algorithms
• Pointing out the information available in the P2P
  resource discovery problem, but which is not yet
  utilized by any local resource discovery
  algorithm
• Reply path forwarding
• Aggregating information from parallel query paths
• Branching factor
• Branching resources to discover

6
Approximation of Optimum for P2P Resource
Discovery Algorithms using k-Steiner Minimum
Trees

• If global information about P2P network is
  available, the optimum for P2P resource discovery
  algorithms can be approximated by solving
  k-Steiner Minimum Tree problem (finding the exact
  optimum would be a NP-complete problem)

7
Approximation of Optimum for P2P Resource
Discovery Algorithms usingk-Steiner Minimum Trees

• MST k-Steiner Minimum Tree Algorithm was
  developed for finding an approximation solution

Time Complexity
Worst-CaseApproximation Ratio
8
Query Path of MST k-Steiner
9
Efficiency

• MST k-Steiner Minimum Tree algorithm (Steiner)
  shows that current local search algorithms for
  peer-to-peer networks are far from optimal

10
Future Work of MST k-Steiner

• The future work of finding optimum consists of
• Getting the results publishedVapa M., Auvinen
  A., Tawast T., Ivanchenko Y., Vuori J.,
  K-Steiner Minimum Tree Is An Upper Bound for
  Peer-to-Peer Resource Discovery Algorithms,
  submitted to IEEE INFOCOM 2006
• Now we have all the tools available for
  discovering the theoretical limit of peer-to-peer
  technology in terms of total traffic induced on a
  telecommunication network in a real-world
  peer-to-peer network compared to client-server
  approach
• Development of distributed k-Steiner minimum tree
  resource discovery algorithm using principles of
  proactive routing protocols such as Open Shortest
  Path First

11
NeuroSearch P2P Resource Discovery Using Neural
Networks

• NeuroSearch resource discovery algorithm uses
  neural networks and evolution to adapt its
  behavior to given environment
• Multiple layers enable the algorithm to express
  non-linear behavior
• With enough neurons the algorithm can universally
  approximate any decision function

12
Performance

• HDS is currently the best known local search
  algorithm for power-law distributed scenario

13
The Swift from Depth-First Search to
Breadth-First Search

• NeuroSearch is close to HDS in performance, but
  different in nature

14
Typical Query Pattern of NeuroSearch
15
Future Work of NeuroSearch

• After two months of extensive simulations with 70
  workstations, we have discovered from 23
  different inputs 7 critical ones (Bias, White,
  PacketsNow, Sent, EnoughReplies,
  FromNeighborAmount and RepliesToGet), which need
  to be present to have good performance
• Next, we are going to boost these 6 inputs by
  generalizing them to give more accurate
  information for forwarding
• Also, we need to discover
• What are the scalability factors of NeuroSearch
  in large graphs
• The performance in dynamic real-world scenarios
  where peers are joining and leaving the network

16
Peer-to-Peer Prototypes
17
Chedar Peer-to-Peer Middleware

• Chedar (CHEap Distributed Architecture) is a P2P
  middleware for searching resources from a
  distributed network
• Resources can be i.e. computing power or files
• Distributed system without any central points
• Contains different resource discovery and
  topology management algorithms
• Implemented with Java 2 Standard Edition

P2P Applications
TCP
Chedar
Chedar
Chedar
TCP
TCP
TCP
Chedar
TCP
IP
Chedar
Network
Chedar
TCP
18
Mobile Chedar

• Mobile Chedar is anextension of Chedarto mobile
  devices
• Bluetooth Java 2 MicroEdition implementationread
  y for SymbianSeries 60
• WLAN Bluetooth Python implementationfor Nokia
  770 Linux Internet Tablet planned for autumn 2005

19
Peer-to-Peer Studio

• P2PStudio is used for measuring the performance,
  visualizing network topology and controlling of
  Chedar peer-to-peer network in an automated and
  centralized manner
• Implemented with Java 2 Standard Edition

Chedar node
Peer-to-Peer Studio
Chedar node
Server
User Interface
Chedar node
Chedar node
20
Peer-to-Peer Distributed Computing

• Peer-to-Peer Distributed Computing (P2PDisCo)
  distributes computations to idling workstations
• Implemented on top of Chedar and deployed in
  Agora building
• The node that offers computation time has to
  implement Distributed interface to be able to
  receive start, stop and is application running
  signals
• Reading of parameters and writing of results are
  done for the stream offered by P2PDisCo
• Any Java program reading input from files and
  writing output to files can be distributed

21
Mobile P2PEncounter Networks

• Information distributes over mobile device
  encounters (Mobile P2P is a future distribution
  model)
• no centralized server, free communication
  bandwidth, no infrastructure
• Applications
• information distribution
• e.g. cheapest bulk product search (gasoline)
• gasoline payment with mobile device
• mobile devices communicate with each other (e.g.
  Bluetooth)
• everybody tells what he/she has paid for the
  gasoline and gets in exchange prices of other
  gas stations
• based on this information, mobile device can
  recommend the cheapest place to fill the tank
• boosts the market based economy by giving equal
  information over the market situation to all
  participants
• grocery store price service, dating service, joke
  service, event service, newspaper service

22
Gasoline Price Comparison System

• Test application for verifying the feasibility of
  mobile peer-to-peer encounter networks using
  Bluetooth
• Uses BlueCheese mobile peer-to-peer middleware
  implemented by the MoPeDi student software
  project during autumn 2003
• Implemented with C for Symbian Series 60 mobile
  devices

GPCS User Interface
23
BlueCheese Protocol Stack
24
DevelopmentHistory and Future
Software
Publications
2002 ----------- 2003 ----------- 2004 -
---------- 2005
Chedar
Data Fusion
P2PStudio
Topology Management

• Research work proceedsas breakthroughs
• P2PRealm network simulator speededup the project
  100x
• P2PDisCo is speeding up the project another 100x
  when fully deployed
• In 2006, the publications side will strengthen
  significantly currently 9 manuscripts are under
  peer review

NeuroSearch
P2PRealm (100x)
Distributed Data Fusion
BlueCheese
NS-2 Simulator
MST k-Steiner
P2PDisCo (100x)
MP2P Co-ope-rative Learning
NeuroSearch
Mobile Chedar
NeuroTopology
Mobile Chedar
Gasoline PriceComparison System
P2PDisCo
Formalization of P2PResource Discovery