Mmogs Application Layer Multicast And Load Balancing Algorithm Research And Implementation

MMOG (Massively Multiplayer Online Game) is a typical class of distributedsystems. Currently, most of the commercial MMOGs are based on client/servercomputing model, which is lack of scalability. Although, server cluster and computinggrid are good ways to improve the scalability of MMOGs, the cost is still too high for thevendors. We can naturally make use of the advantage of scalability of Peer-to-peernetworks to build MMOGs systems. P2P (Peer-to-Peer) networks are being used innearly all of the networking applications, some of which typically includes file sharing,scientific computing, multi-media applications and so on. In comparison of traditionalC/S computing model, P2P weakens or evenly eliminates the functionality of centralservers. It can fully take advantage of the resources on the hosts distributed in thenetworks, namely edge-resources, and organize them by flexible cooperated structures. Intheory, the topologies, formed by P2P networks, are better at scalability than that of C/Smodel. According to the researching project "MMOGs Based on P2P Network andRelated Technology" funded by Natural Science Fund in Liaoning Province, we designeda new generation of online gaming engine, called Rambler. The work in this thesis isbased on Rambler. Two important problems are discussed in this work, application layer multicast andload balancing in P2P networks based on DHT. Comparisons and analysis on relatedtraditional algorithms are accomplished in this work. According to the results of formerwork and the characteristics of MMOGs, a multicasting node's joining algorithm and adynamic load balancing algorithm for MMOGs are presented in this work. Firstly, thiswork decouples the communication domain of MMOGs into two layers, namelywithin-AoI and inter-AoI. On within-AoI, a new multicasting node's joining algorithm isproposed. On inter-AoI, this work suggests and discusses a multicasting tree creatingalgorithm. There are two layers of load balancing problems, and in the first one, anexisted ID insertion load balancing algorithm is suggested and discussed. In the secondone, this work presents a new greedy dynamic load balancing algorithm. According to theexperiments' results, both of the two algorithms presented in this work are feasibleenough for Rambler.