Towards Reinterpretation of Interaction Complexity for Load Prediction in Cloud-based MMORPGs

1
Towards Reinterpretation of Interaction
Complexityfor Load Prediction in Cloud-based
MMORPGs

• Mirko Sužnjevic, Maja Matijaševic

University of Zagreb Faculty of Electrical
Engineering and Computing Department of
Telecommunications
http//www.fer.unizg.hr/ztel/en/research/research_
groups/netmedia
2
Motivation
Number of players on a server 4000
Number of players on a server 4000
Not the same in terms of load!
3
Outline

• Introduction
• Related work
• Mapping
• Load prediction
• Resuts
• Conclusion and open issues

4
Introduction

• Current practices in server organization for
  MMORPGs
• Sharded worlds (World of Warcraft)
• Single worlds (EvE Online)
• Problems
• Player distribution uniform (in specific shard
  and between shards)
• Player behavior patterns
• Unexpected behavior
• Highly variable load
• Cloud based systems designed to provide optimal
  amount of resources for services with variable
  load

5
Related work

• Based on work by Nae and Prodan
• Presented formulas for load estimation (both CPU
  and memory) in distributed MMOs
• Created load prediction based on neural networks
• Used Runescape as a case study
• Defined complexities for different interactions
  between entities in MMOs
• We map the MMORPG behavior categories to specific
  complexities
• R. Prodan and V. Naea, Prediction-based
  real-time resource provisioning for massively
  multiplayer online games, Future Generation
  Computer Systems, vol. 25, pp. 785793, 2009.
• V. Nae, A. Iosup, and R. Prodan, Dynamic
  resource provisioning in massively multiplayer
  online games, IEEE Transactions on Parallel and
  Distributed Systems, vol. 22, pp. 380395, 2011.

6
Mapping

• Goal - to prove that behavioral information has a
  significant influence on the system load
• Redefined calculation of number of interactions
  based on properties of the action category

7
Load prediction

• Simple history based algorithm
• History source previously developed user
  behavior simulator
• Behaviour simulator implements the model which
  captures behavior patterns
• Easy modification of parameters of the behavior
  (number of players, duration of each separate
  action category, arrival and departure rates)
• Enables simulation of extreme events such as
  flash mobs and content release
• Output in form of a text file listing each
  performed behaviour Second124 TypeChange
  ID210 Category5 CategoryOld6 Duration433

8
Load prediction II

• Developed a tool for load prediction
• Run by simple algorithm based on pattern periods
  (e.g., day, week)
• The history is divided to intervals (e.g.,
  minutes, hours)
• It takes into account the whole history and last
  period in order to capture both patterns and
  unusual behavior

Second124164 TypeChange ID2610 Category5
CategoryOld6 Duration433
9
Results

• Tool displays 4 curves
• Whole history
• Last period
• Predicted resources
• Reserved resources
• Two calculation algorithms
• Taking into account only number
• of players
• Taking into account both number
• of players and behavior category

Second124164 TypeChange ID2610 Category5
CategoryOld6 Duration433
10
Conclusions and open issues

• Conclusions
• Adapted the interaction complexities to specific
  action categories
• Developed a load prediction tool
• Confirmed significant discrepancies when
  calculating load just based on the number of
  players and based on behavioral information
• Open issues
• Issues which cloud based MMORPGs face?
• More interaction between shards
• Is just adding another server when load hits
  certain treshold enough?
• Load prediction based on spatial locality of the
  players behavioral information

11
(No Transcript)