Fault-tolerance for massive-scale distributed systems

云计算既指通过互联网以服务形式提供的应用程序，也指提供这些服务的数据中心的硬件和系统软件。随着亚马逊(Amazon Inc.)和谷歌(Google LLC)等互联网公司的崛起，过去十年云计算的使用迅速增加。这些基础设施以现收现付的方式与提供基础设施即服务(IaaS)的公司合作，允许客户根据其应用程序需要设置和定制执行环境。随着计算机硬件的速度和能力的增加，视频游戏开发商仍在挑战这种硬件可以执行的极限。然而，随着电力的增加，价格也随之上涨，玩家往往负担不起建立一个能够在最高设置下运行游戏的系统。这就是云计算与游戏相结合，提供按需游戏流媒体的地方。目前的方法，如Outatime和Nvidia Grid，允许在多个云服务器上分发完整的游戏，并将结果回流到最终用户。有了这种设置，用户不再需要拥有一台功能强大的机器就可以运行游戏。由于客户端现在需要从云提供商那里租用游戏，但由于服务器的硬件要求提高了租金，因此仍然可能很昂贵。近年来，在线多人游戏也变得占主导地位，在流行的PC平台上玩得最多的十款游戏中，有七款要么主要是在线游戏，要么完全是在线游戏。对于在线多人游戏，由于同步依赖关系，许多交互模拟必须同时在服务器端和客户端执行。对于流媒体在线游戏，这种重复的计算可以完全减轻，因为游戏状态完全在服务器上，服务器之间只需要来自客户端的输入就可以进行通信。与分发整个游戏不同，这项研究将调查以云爆裂的方式分发游戏引擎的计算繁重部分的可行性，以保持游戏在重负载下的性能。这将降低服务器的资源需求，因为引擎的部分仍将由客户端计算，如果它们的计算机足够强大的话。

Cloud Computing refers to both the applications delivered as services over the Internet and the hardware and systems software in the data centres that provide those services. The use of cloud computing has rapidly increased over the last decade with the rise of internet corporations such as Amazon Inc. and Google LLC. These infrastructures work on in a pay-as-you-go manner with companies providing Infrastructure as a Service (IaaS) allowing clients to set up and customise execution environments to their applications needs.As the speed and power of computer hardware increases, video game developers are still pushing the limits of what this hardware can execute. However, along with this increase in power comes an increase in price and often players cannot afford to build a system capable of running the games on the highest settings. This is where cloud computing can be combined with gaming to provide on-demand game streaming. The current approaches, such as Outatime and Nvidia Grid, allow for full games to be distributed across multiple cloud servers with the results being streamed back to the end-user. With this setup the user is no longer required to own a powerful machine so that they can run the game. As the client is now required to rent games from the cloud providers it can still be expensive due to the hardware requirements of the servers increasing the rent price.Online multi-player gaming has also become dominant in recent years, with seven of the top ten most played games on the popular PC platform 'Steam' being either mostly online or fully online. With online multiplayer games a lot of the interactive simulation must be performed both server side and client side due to synchronisation dependencies. With streamed online games this duplicated computation can be mitigated entirely as the game state is entirely on the servers which can communicate between themselves only requiring input from the clients.Instead of distributing entire games this research will investigate the viability of distributing computationally heavy sections of the game engine in a 'cloud bursting' style to keep the game performant under heavy load. This will lower the resource requirements for the servers as sections of the engine will still be computed by the client if their machine is powerful enough to do so.