Directed Research Project Details
PopCorn: Placement of Point-of-Presence Applications in Open Peer-to-Peer Environment
| Author: | Yuhong Liu |
|---|---|
| Date: | March 11, 2005 |
| Time: | 10:00 |
| Location: | 220 Deschutes |
| Committee: | Virginia Lo (Chair) Arthur Farley Stephen Fickas |
Abstract
Point-of-Presence (PoP) applications are one class of applications that need to disperse a set of tasks throughout the Internet to collect and process data. The current approach for placement of PoP tasks is to manually select a group of hosts according to the specific requirements of the PoP application and to manually deploy tasks on those hosts. However manually placement may lead to poor dispersal and is not possible for large scale overlay network. I developed PopCorn, a fully decentralized protocol that disperse tasks throughout the overlay using a repulsion model among task tokens. My simulation demonstrates the effective of PopCorn by measuring ATT (Average Token-to-Token distance) and ATN (Average Token-to-Nontoken distance). Also I analyze the communication overhead and convergence time of PopCorn. The simulation results show that PopCorn, with respect to dispersal metrics, is close to that of Gonzalez's centralized greedy heuristic, and is scalable with low message overhead. PopCorn performs better in powerlaw graphs than in transit-stub graphs.