標題: 雲端大型多人線上遊戲下基於玩家行為之資源分配
Player Behavior-based Resource Allocation for MMOG Clouds
作者: 賴寬嶧
Lai, Kuan-Yi
王國禎
Wang, Kuo-Chen
資訊科學與工程研究所
關鍵字: 雲端計算;大型多人線上遊戲;玩家行為;資源分配;cloud computing;MMOG;player behavior;resource allocation
公開日期: 2013
摘要: 現今的大型多人線上遊戲(MMOG)已有超過數以千萬的用戶,其中較受歡迎的遊戲可能有超過一萬人同時在線上。因此,為了解決遊戲伺服器所產生的大量負載變動,許多大型多人線上遊戲的營運商,企圖將他們的遊戲服務放到雲端平台上運行,以便善用雲端計算的優點。遊戲伺服器的資源 (CPU、記憶體、網路頻寬) 需求量和玩家的關注區域 (Area of Interest) 內有多少玩家,以及該玩家正在做什麼的行為有很大的關係。資源分配不足 (under-allocation) 會導致玩家好的遊戲體驗下降,使得玩家離開遊戲並刪除帳號。為了保證玩家有好的遊戲體驗,現在大型多人線上遊戲的營運商大多都是採用超額配置 (over-allocation) 資源這種策略。然而,過度的超額配置資源會導致資源整體的使用率下降。為了解決這個問題,我們提出一個雲端大型多人線上遊戲下基於玩家行為之資源分配 (PB-RA)方式。我們透過類神經網路來預測未來地圖上各種行為的玩家人數,根據量測不同玩家行為所產生的負載給定一個更精確的伺服器整體資源需求量,讓我們在分配資源時可以更有效率。實驗結果證明,我們所提出的基於玩家行為之資源分配方式,相對於只考慮玩家人數來分配資源的方法減少74%的資源超額分配,相對於考慮玩家互動來分配資源的方法減少50%的資源超額分配。此外,資源分配不足發生的次數相對於考慮玩家互動來分配資源的方法也不超過1.05倍。據我們所知,目前並沒有大型多人線上遊戲的資源分配方式有考慮到玩家行為。
Today's Massively Multiplayer Online Games (MMOGs) have more than tens of millions subscriptions and the popular one may have over 10,000 active concurrent players. Therefore, many MMOG operators attempt to run their game services in clouds to take advantages of cloud computing, such as on-demand self-service and resource pooling characteristics, in order to handle large load variation in game servers. The amount of resource (CPU, memory, and network bandwidth) requirements for a player is related to how many players are in his Area of Interest (AoI) and what kind of player behavior. Resource under-allocation leads to degradation of game experience of players and may trigger player quitting and account closing. To guarantee the better players’ experience, resource over-allocation is the most commonly used resource allocation policy adopted by MMOG operators. However, resource over-allocation often results in low overall resource utilization. To address this deficiency, we propose a dynamic Player Behavior-based Resource Allocation scheme for MMOG clouds, called PB-RA. We predict the number of players for each behavior type in one map through a neural network-based predictor, and measure loads generated from different player behavior types. As a result, we can predict total resource requirements more accurately for players with different behavior types in the map. That is, we can allocate resources more efficiently. Experiment results show that the proposed PB-RA can reduce 74% and 50% of resource over-allocation compared to the method that only considers number of players and the method that considers interaction of players as well, respectively. Moreover, in terms of the number of resource under-allocation events, the proposed PB-RA is no more than 1.05 times compared with the method that considers interaction of players. To the best of our knowledge, there is no resource allocation scheme for MMOGs that considers player behavior.
URI: http://140.113.39.130/cdrfb3/record/nctu/#GT070056095
http://hdl.handle.net/11536/72237
Appears in Collections:Thesis


Files in This Item:

  1. 609501.pdf