You are invited to attend Jason Long's Master's Thesis defense on Friday, September 7th at 2pm. The defense will be held in Olmsted W231. Details about Jason's thesis can be found below.
Title: SALIENT: Stochastic, Adaptive Latency Improvement for Event Notification Trees
Author: Jason Long
Thesis Adviser: Jeremy Blum, D.Sc.
Abstract
A challenge in massively multiplayer online games (MMOGs) is the need for game event information to be quickly disseminated to all participants. Because of the cost and scalability limitations of centralized servers, peer-to-peer technologies have been adopted in which peers serve both to reconcile conflicting actions and to broadcast the events to other peers. This paper describes and evaluates a new algorithm, Stochastic, Adaptive Latency Improvement for Event Notification Trees (SALIENT). SALIENT provides for the efficient broadcast of event dissemination in peer-to-peer MMOGs through a distributed algorithm for constructing and maintaining a peer-to-peer broadcast tree.
SALIENT consists of am MMOG event dissemination model requiring a zone coordinator to be chosen from all participants. Each participant sends their event data to that zone coordinator for the zone coordinator to serialize and then broadcast through a broadcast tree to all participants.
The aim of SALIENT is to decrease the experienced average latency of all participants. In order to accomplish this goal, SALIENT employs a number of independent mechanisms that work over time to incrementally make improvements to the structure of the broadcast tree. SALIENT incorporates randomized algorithms to continually explore alternate tree structures that might reduce average latency and evaluate those alternate tree structures. When an improved structure is found, SALIENT adopts the new structure.
SALIENT was found to greatly reduce experienced latency, typically as much as 30-50% reduction in event delays, as compared to a simple base case where the broadcast tree is randomly assembled. In addition, SALIENT has bandwidth management strategies that help increase the number of nodes that can be supported for nodes with very limited bandwidth. Compared to a base case where a flat broadcast tree is used and can only support 12 participants in total, SALIENT allows 60 or more participants to be supported.
