Prediction-based real-time resource provisioning for massively multiplayer online games

  • Authors:

  • Radu Prodan;Vlad Nae

  • Affiliations:

  • Institute of Computer Science, University of Innsbruck, Technikerstraíe 21a, A-6020 Innsbruck, Austria;Institute of Computer Science, University of Innsbruck, Technikerstraíe 21a, A-6020 Innsbruck, Austria

  • Venue:
  • Future Generation Computer Systems
  • Year:
  • 2009

Quantified Score

Hi-index 0.00

Visualization

Abstract

Massively Multiplayer Online Games (MMOGs) are a class of computationally intensive client-server applications with severe real-time Quality of Service (QoS) requirements, such as the number of updates per second each client needs to receive from the servers for a fluent and realistic experience. To guarantee the QoS requirements, game providers currently over-provision a large amount of their resources, which makes the overall efficiency of provisioning and utilization of resources rather low and prohibits any but the largest providers from joining the market. To address this deficiency, we propose a new prediction-based method for dynamic resource provisioning and scaling of MMOGs in distributed Grid environments. Firstly, a load prediction service anticipates the future game world entity distribution from historical trace data using a fast and flexible neural network-based method. On top of it, we developed generic analytical game load models used to foresee future hot-spots that congest the game servers and make the overall environment fragmented and unplayable. Finally, a resource allocation service performs dynamic load distribution, balancing, and migration of entities that keep the game servers reasonably loaded such that the real-time QoS requirements are maintained. Experimental results based on a realistic simulation environment demonstrate the advantages of our prediction service compared to other conventional methods, especially due to its ability to adapt to different user load patterns, and a reduction of the average over-allocation from 250% (in the case of static over-provisioning) to around 25% using our dynamic provisioning method.