Simulation of Partition-resilient Overlay Peer-to-Peer Networks

Project Details

Project Lead
Renato Figueiredo 
Project Manager
Renato Figueiredo 
Project Members
María Madrid Pérez, Gicu Michiu  
Institution
University of Florida, Electrical and Computer Engineering  
Discipline
Computer Science (401) 
Subdiscipline
11.01 Computer and Information Sciences, General 

Abstract

This project investigates the resiliency of Peer-to-Peer (P2P) overlay networks - based on both structured topology as well as a social network-based topology - against partitioning attacks targeting a relatively small fraction of the overlay network (e.g. 0.5% of the network is "cut"). An example of this might be a Government that disconnects the Internet in the country from the rest of the world (while allowing networking inside). FutureGrid cloud and if needed large-memory resources will be used for graph simulation and evaluation.

Intellectual Merit

First, we would like to investigate the resiliency of Peer-to-Peer (P2P) Structured Overlay Networks (SON) against partitioning attacks targeting a relatively small fraction of the overlay network (e.g. 0.5% of the network is "cut"). An example of this might be a Government that disconnects the Internet in the country from the rest of the world (while allowing networking inside). What fraction of users in that country would still be able to communicate among each other, after the partitioning attack? How worse does it get when a significant fraction of the nodes are behind different NATs? And what about if the adversary colludes with a small fraction of the users within the small disconnected component? Furthermore, we want to evaluate this hypothesys: is a Unstructured Overlay Network (UON) based on social relations more resilient than a SON, against the attacks described above?

Broader Impacts

This project has potential broader impact on peer-to-peer systems that are able to better maintain connectivity among users in the face of large network partitions.

Scale of Use

The anticipated scale of use is a few VMs (typically one) with relatively large memory (16GB+) deployed for a few hours at a time.