未来网络技术项目白皮书

王永浩 于 20129

国际互联网的困境

当今,因特网上大部分的数据流量是实时音频和视频内容,例如流媒体,网络电视等。普通数据传输 例如电子邮件和文件传输反而占了相对较少的部分。 然而因特网从设计之初就在实时信号传输上有缺陷(用户在使用网络语音通信,网络视频等的不稳定的体验),更不用说支持高清的交互式多媒体应用和其他对实时 性要求很高的商业应用了(例如跨洲际高频交易和电子银行业务等)。

当今科技界缺乏一种网络架构可以同时高效的支持实时流量(Real-Time Data),又可以尽可能的有效的支持普通数据业务(Best-Effort data)。设计这样的网络架构也有很大的技术难度。长期以来,使用TCP/IP架构的网络被认为是可以支撑多网合一的综合业务,并且有很好的地域扩展性。但 是基于非面向连接的包交换架构的IP网络对支撑“低延时实时数据”业务不可避免的存在着先天技术缺陷。现有的解决方案,例如复杂的服务质量(QoS)管理,流量工程技术,以及冗余带宽方案都不能从根本上解决问题,相反,他们大大增加的系统的复杂度,成本和能耗需求。而且还不能给用户以百分之百的服务质量需求。

我们的技术

我们提出一种新型的网络架构叫做(AudioVisual-Friendly IPAVFIP,可以有效的支持实时数据流和普通(Best-Effort)数据流,并充分的利用有效带宽,同时设计了高性能的交换器架构。对于实时性要求非常高的应用,例如交互式多媒体,语音视频通信等,AVFIP能够同时确保低延时的性能以及灵活的支持利用冗余带宽传输其他非实时的数据流。AVFIP架构并且能够支持多路通道的,完全独立的,不同速率的实时数据流,无论其实时数据流是定长包结构还是变长包结构,其网络带宽都能得到非超好的利用,并且保证用户的QoS体验。

开发进度

现阶段该技术已经进入原型机和小规模实用阶段,点到点通信的网卡原型已经完成。使用成本只有10美金的FPGA硬件就可以实现新型网络协议。高精度实时数字音频交换机的原型机也将于2012年底完成。届时可以演示高清晰,低延时的多轨音频在拥塞网络上运行的性能。

物理层:对以太网上跑AVFIP的研究已经基本完成。下一步是对光纤网络和无线网络的研究,给AVFIP应用在无线网络和长距离通信上打下基础。

逻辑链路层:已经完成专利技术Flexilink的设计和实现,其通信控制面基于自主开发的流处理器(Flow Processor)架构和网络编程语言,处于世界领先地位。

网络层和高级应用层:对现有IPv4IPv6的兼容在体系架构上已经实现。需要进一步开展软件驱动的开发工作,一旦完成,AVFIP将对现有的网络媒体应用起到革命性的性能提升作用。



核心技术优势



业内同行评议

世界领先的地位和渠道

The research project is based on a collaboration on “Future Network” and its potential applications with John Grant of NineTiles Networks who is a leading editor of ISO/IEC standard drafts on related areas. The “Future Network” project is based on a clean-slate design of future Internet infrastructure in order to overcome some inevitable systematic problems of the current Internet architecture.

Current Development

Currently the novel Layer-2 architecture Flexilink is proposed. The initial work will be presented in 133rd AES convention [1]. The ultimate aim of this layer-2 redesign is:



·        Fully manageable and measurable QoS for time critical data such as Audio/Video streaming, interactive media.

·        Low jitter, low latency, no packet loss for time-critical data.

·        Towards full bandwidth utilization. (Less over-provisioning or wasted capacity) 

The design rationale and concept are based on a simplified taxonomy of Internet traffic and the redesign of layer 2 architecture, the details of which can be found in[1]. In this initial work the implementation of Flexilink on Gigabit Ethernet is discussed and proposed. However, there are many unsolved areas and problems such as the design of Flexilink over different layer 1 network medium especially wireless and mobile network, and the interconnection of Flexilink with existing network architectures.

 

Possible research roadmap

An academically rigorous theoretical model needs to be established to quantify and evaluate the quality of the new architecture. It is necessary to model the new architecture “Flexilink” using network simulation tools. And it is interesting to see how Flexilink works over other existing network technology. The following topics need to be further investigated as potential research projects:

·        Simulation of Flexilink on OPNET, NS2 or NS3.

·        Flexilink over different layer1-2 technology PON, DSL, Wireless.

·        IP and Flexilink, Routing, multicasting and unicasting, MPLS and Integrated service.

·        Router Switch design (FPGA + Network Flow processing language)

Application Projects

It is also foreseeable to have some application oriented projects to align with the major subject area of the School of Digital Media Technology such as:

·        Studio networking (Audio/Video over Flexilink).

·        Low latency remote haptic over network.

·        Synchronized Cloud collaboration.

·        CDN Video networking delivery.

Reference

[1] Y. Wang, J. S. Grant, and J. Foss, ‘Flexilink: A unified low latency network architecture for multichannel live audio’, presented at the 133rd AES (Audio Engineering Society) Convention, San Francisco, CA, USA, 2012.