利用圖形處理器加速鉅量多人連線遊戲伺服器端之運算

Abstract

一般來說，MMOG（鉅量多人連線遊戲）主要包含了一個廣大的虛擬世界，容許成千上萬的玩家在這個世界裡進行遊戲以及互動。然而，當我們要設計一個MMOG的平台（或中介軟體）來協助遊戲的開發時，我們常常會碰到各式各樣的瓶頸，限制了在單一平台上最多能容許的玩家人數。本研究探討了這限制之最根本的原因，就是在CPU上循序的處理遊戲邏輯；我們提出了一個革命性的方法，讓GPU（圖形處理器）來加速這些遊戲邏輯的計算，減低CPU的負擔，讓CPU更專心於處理OS以及I/O方面的訊息。 本論文是目前為止第一個利用了GPU的計算能力，來加速MMOG伺服器端之運算的研究。我們提出了數個平行的演算法，將用來原本該在CPU上之邏輯運算，轉換成可以在GPU上平行處理的演算法。在我們的實作中，隨著玩家人數的增加，我們觀察到在GPU提供了超過100倍目前CPU所能提供之運算能力。搭配適合的IO架構，在單一伺服器上服務超過50萬個玩家即可被實現。

In general, MMOG (Massively Multiplayer Online Game) consists of virtual worlds where thousands or millions of people can login into the same place and interact with each other. However, when designing a generic MMOG platform to support game development, there are always bottlenecks which limit the maximum number of concurrent players within a single server cluster. In this paper, we identify the source of the limitation, which comes from sequential processing overhead, and propose an innovative approach to solve the scalability problem in parallel by GPU (Graphics Processor Unit). This is the first research that exploits GPU computation power to accelerate MMOG server-side computing. Several parallel algorithms handling server-side computation of MMOG are derived and implemented on GPU. In our implementation, we observed the performance improvement of GPU is more than 100 times, compared to modern CPU. Together with underlying supporting network communication infrastructure, we build a scalable and high performance MMOG platform that serves up to 500k players on line with a single server at low cost.