基於熱點預知之雲端大型多人線上遊戲之動態資源配置

Abstract

近年來，有許多類型的遊戲改為網路服務的經營模式。大型多人線上遊戲(MMOG)成為世界上最受歡迎的遊戲服務類型。為了節省營運成本及簡化遊戲伺服器的管理，有許多遊戲服務供應商將他們的服務與雲端計算科技結合。在大型多人線上遊戲的虛擬世界中，玩家經常藉由成群結隊的行動，來達成某些遊戲任務或擊敗遊戲中的魔王。這樣的行為模式可能造成虛擬世界中的＂熱點＂。在熱點中的玩家，常需頻繁地互動，產生大量的負載，以至於造成服務品質的下降。若有太多的熱點同時存在同一台伺服器中，遊戲的流暢度將會受到影響。為了解決這個問題，傳統的方式會藉由高估單一遊戲地圖的負載量，分配遠超過需求的資源來維持遊戲品質，但此會造成遊戲資源上的浪費。在本論文中，我們提出一個基於熱點預知的動態資源配置方案(NN-Player+DRP-HA)，並且使用一個有限狀態機來表示玩家狀態及其可能的狀態轉換。我們結合玩家的狀態和類神經網路(neural network)對下一個時間點玩家數量進行預測，我們可以計算出地圖上熱點造成的潛在負載，並且分配適當的計算資源來消化這些負載。實驗結果顯示，我們提出的方法可以在不造成嚴重的資源過度配置的情況下，降低資源配置不足的機率。與現存的另一代表性之動態資源配置方法(NN-Player+DRP)比較，我們提出的方法可以以控制CPU過度分配的比率不超過一台虛擬機器容量的前提下，將CPU資源分配不足的次數從2.16% 降低至0.42% (改進了80%)。

Recently, there are various kinds of games that have been served via the internet. An example of such games is MMOG (Massively Multiplayer Online Game) that has become the most popular game service in the world. For saving operating cost and simplifying management of servers, gaming service providers are combining their online services with cloud computing technology. In MMOG virtual environments, avatars are often acting as a group to help one another to achieve certain goals or defeat bosses, which may become a hotspot in a virtual world. Frequent interactions between avatars in hotspots may generate lots of workload and may cause decrease of quality of service (QoS). The latency of gaming service will increase when there are too many hotspots in a single server, which may harm the quality of experience (QoE) enormously. To address this problem, in general, games operators over-allocate resources to game zones, which may cause the waste of gaming resources. In this paper, we propose a dynamic resource provisioning with hotspot anticipation scheme, called NN-Player+DRP-HA that employs a vector based model to monitor the movement of avatars in a virtual world. Furthermore, we use a finite state machine to represent possible avatar states and state transitions. By combining the state of each avatar in a game zone with a neural network (NN) predictor, we may figure out potential workload produced by hotspots, and then allocate appropriate computing resources to support the game zone. Experimental results support that the proposed NN-Player+DRP-HA scheme can avoid most of under-allocation events with an acceptable over-allocation rate. Compared with a representative dynamic resource provisioning method, called NN-Player+DRP, the proposed NN-Player+DRP-HA reduces the probability of under-allocation events from 2.16% to 0.42% (80% improvement) in terms of CPU capacity of a VM, under the premise of controlling the CPU over-allocation rate within the CPU capacity of one VM.