background image
Scalable Agent-Based Simulation of Players
in Massively Multiplayer Online Games
Amund Tveit
Øyvind Rein
Jørgen V. Iversen
Mihhail Matskin
Department of Computer and Information Science,
Norwegian University of Science and Technology
N-7491 Trondheim, Norway
Abstract.
We propose a parallel mobile agent platform - Zereal - for scalable and flex-
ible simulation of Massively Multiplayer Online Games (MMOGs). Players and NPCs
in Zereal currently have a sense-reason-act behavior, with reasoning based on Markov
chains in addition to hierarchical plans specified in XML. Zereal's primary purpose
is to be a MMOG simulation tool that enables testing of usage logging approaches
for CRM and data mining purposes, and secondly to allow flexible testing of behav-
ioral AI models for players and non-personal characters. Zereal has shown to be close
to linearly scalable in terms of number of players, and has been successfully tested
with more than 100 thousand players on 20 CPUs on a Linux-based cluster. Zereal is
implemented using Python, MPI and C++.
1
Introduction
Over the last couple of years an emerging commercial online games market has come into
existence, fuelled by the increasing availability of cheap Internet access. The main share of
this market consists of Massively Multiplayer Online Games (MMOGs). The characteristic
property of MMOGs is that a large number of players, sometimes even in the six or seven
figure realm, take part in a persistent virtual world where they communicate, cooperate, fight,
build virtual characters, and undertake game-related missions [11, 24].
The worldwide MMOG market is anticipated to grow annually by 70% to USD 2.7 bil-
lion in year 2006 [12]. Traditionally the game industry has been highly product-focused, but
with the introduction of MMOGs games have become more like services, where players not
necessarily pay for the MMOG game client, but instead pay a monthly subscription fee for
having access to the MMOG world [8, 15]. In order to gain and keep subscribing players there
is a need to focus on Customer Relationship Management (CRM) and Data Mining methods
[1, 21]. The overall motivation for such methods in a MMOG context is to figure out more
about the players and their preferences so the MMOG can be continously improved and keep
on generating revenue.
1.1
Research Problem and Approach
A fundamental requirement for data mining and CRM methods is the need for relevant usage
data
. For MMOGs there are currently no open standards for usage data logging, the data is
80
Scalable MMOG Player Simulation

<< - < - > - >>