基于云计算平台的网络虚拟环境关键技术研究

With the development of networking and virtual reality techniques, Networked Virtual Environments (NVEs), represented by online games and online conferences, are becoming one important type of Internet applications with a large number of users and enormous economic value. The number of NVE user is massive and fluctuates over time, and the NVE user behaviors are numerous and dynamic. These bring great opportunities and challenges to distributed computing community. Researchers have proposed various techniques which can be used to increase the scalability of NVEs and reduce their operational cost. However, these challenges are only partially solved, for example, the event-based lockstep simulation technique can significantly reduce the network consumption of NVEs, but it has high CPU consumption. In this proposal, first of all, we collect and study large-scale user behaviors and system workloads to understand their patterns. Then, based on the analyzed patterns and traditional scalable techniques, we propose an area-of-simulation based scalable technique to support the numerous and dynamic behaviors of large-scale users. At last, by making use of the different properties of various computing resources and pricing options of cloud computing resources, we propose a resource scheduling technique which can serve a dynamic and a huge number of users with low-cost. The scalable technique and the resource scheduling technique can be combined effectively, to solve the mentioned challenges, and have a great impact on NVEs, which have large user-base and huge market value.

随着网络及虚拟现实技术的发展，以网络游戏、网络会议为代表的网络虚拟环境已成为一类重要的互联网应用，有着显著的用户及经济效益。网络虚拟环境用户数量多且多变、其用户行为多且多变，这为分布式计算带来极大的机遇与挑战。学术界对此有广泛兴趣并提出了不同可扩展及资源调度方法以可扩展、高效地服务行为多且多变的海量用户。但是这些方法只解决了部分挑战，比如事件同步模拟技术能有效降低网络开销，但CPU开销较高。本项目首先测量分析网络虚拟环境海量用户行为及系统负载，深刻理解其规律；其次，结合发现的规律及传统可扩展技术的优点，设计可扩展的区域模拟技术以支持海量用户多样多变的行为；最后，充分利用云计算平台不同计算资源及计费模式的特点，设计云资源调度技术以较低成本服务海量且多变的用户。本项目提出的可扩展技术及云资源调度技术可以有机地结合起来一起解决上述挑战，将对拥有巨大用户基数及重要产值的网络虚拟环境产生重大影响。

随着网络及虚拟现实技术的发展，以网络游戏、网络会议、社交视频，社交直播为代表的网络虚拟环境已成为一类重要的互联网应用，有着显著的用户及经济效益。网络虚拟环境用户数量多且多变、其用户行为多且多变，这带来极大的机遇与挑战。学术界对此有广泛兴趣并提出了不同可扩展及资源调度方法以可扩展、高效地服务行为多且多变的海量用户。但是这些方法只解决了部分挑战，比如事件同步模拟技术能有效降低网络开销，但CPU开销较高。本项目的研究内容包括三个方面：(1) 测量海量网络虚拟环境用户行为及系统负载，利用云计算资源采集了多个网络虚拟环境多达上百万虚拟用户数据以及上千万的用户行为数据。基于采集数据进行分析，从而深刻理解用户行为、用户数量多及多变的特性及规律。(2) 对于用户行为多样多变及数量多的挑战，本项目结合区域兴趣及事件同步模拟机制的优点，提出了可扩展的网络虚拟环境区域模拟技术以支持海量用户多样多变的行为，能够以较少的资源消耗支持比传统技术多一个数量级的用户。(3) 对于用户数量多及多变的挑战，提出了基于公文包调度的云计算资源调度技术以服务网络虚拟环境用户并显著降低计算资源使用成本。