Skip to content

The goal of this project was to determine the convergence of Gossip type algorithms through a simulator based on actors written in Elixir. Full, Line, 2D and Imperfect 2D topologies were implemented for both Push Sum and Gossip algorithms. We could demonstrate convergence of these algorithms with nodes upto 80000 without hitting system limits. T…

License

Notifications You must be signed in to change notification settings

anikethsukhtankar/gossip-push-sum-protocol

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

7 Commits
 
 
 
 
 
 
 
 
 
 

Repository files navigation


DISTRIBUTED OPERATING SYSTEM - GOSSIP SIMULATOR

CONTENTS OF THIS FILE

  • Required Answers
  • Introduction
  • Requirements
  • Installation and Configuration
  • Program flow
  • Developer information

REQUIRED ANSWERS

  • Team members

    Shikha Mehta (UF ID 4851 9256)

    Aniketh Sukhtankar (UF ID 7819 9584)

  • What is working

    • 100% convergence is being achieved each time for Gossip in all the topologies. The upper limit for the number of nodes that can be handled by the system is only due to the system limits on the number of processes.
    • 'Line' topology takes a large time to converege for gossip and push sum while 2d takes over line for large networks.
    • Moreover, Gossip is working for Full, 2D and Imperfect 2D for nodes upto 80000 without hitting system limits and converges better than line.
    • Pushsum took much longer to converge when compared to gossip as the sum from one end of the network had to reach the other end of the network which was not pssobile in many cases (large networks and line topology). Deadlocks were encountered while running push sum for line topology and large nodes,
    • In conclusion, All the Topologies have been implemented for both Push Sum and Gossip algorithm. I have also implemented a failure model and tested it with multiple parameters (Number of killed nodes, Number of nodes triggered at initiation and Conergence criteria). I have tested the failure model on all the topologies as well as both Push Sum and Gossip algorithms.
    • The advantages of the gossip algorithms are that they are completely decentralized, have excellent fault tolerance, are Iterative and probabilistic which makes them imperfect. The algorithms run fast in certain scenarios and slow in others. The Push Sum algorithm is good for computing sum and can be extended to compute many types of linear functions, random sampling, quantiles, etc.

INTRODUCTION

The project folder contains 2 folders:

  • gossip: Contains the folders related to the gossip simulator project. Important files being: - Main.ex - Implementation.ex - Gossip and Push Sum Actors - report.pdf

  • bonus: Contains the folders related to the bonus project. - Main.ex - Implementation.ex

    • Gossip and Push Sum Actors
      • bonus-report.pdf

REQUIREMENTS

The following need to be installed to run the project:

  • Elixir
  • Erlang

INSTALLATION AND CONFIGURATION

Go to the folder 'project2' using command line tool and type escript project2 numnodes topology algorithm.

PROGRAM FLOW

  • GOSSIP

    • enter the number of nodes (will be rounded up for 2d and imp2d topologies)
    • enter the topology
    • enter the algorithm
    • (optional) enter the number of nodes triggered at initiation
    • (optional) enter the stopping criteria (eg. 100 if there are a 1000 nodes in the network and a convergence rate of 90% i.e. 900 nodes converging is good enough)
  • BONUS

    • enter the number of nodes (will be rounded up for 2d and imp2d topologies)
    • enter the topology
    • enter the algorithm
    • enter the number of nodes to kill (used this feature for testing and analyzing various topologies and algorithms)
    • (optional) enter the number of nodes triggered at initiation
    • (optional) enter the stopping criteria (eg. 100 if there are a 1000 nodes in the network and a convergence rate of 90% i.e. 900 nodes converging is good enough)

DEVELOPER INFORMATION

Aniketh Sukhtankar (UF ID 7819 9584) asukhtankar@ufl.edu

Shikha Mehta (UF ID 4851 9256) shikha.mehta@ufl.edu

About

The goal of this project was to determine the convergence of Gossip type algorithms through a simulator based on actors written in Elixir. Full, Line, 2D and Imperfect 2D topologies were implemented for both Push Sum and Gossip algorithms. We could demonstrate convergence of these algorithms with nodes upto 80000 without hitting system limits. T…

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages