Smoothing and Congestion Control Algorithms for Stored Continuous Media

未来的高速计算机网络预计将承载一系列令人眼花缭乱的网络多媒体应用，包括数字图书馆、视频和图像服务器、远程学习和协作、网络虚拟环境和娱乐。 这些应用程序之间的一个共同点是其数据的所谓“连续”性质。 在连续媒体（CM）数据中，存在严格的定时关系，其定义CM数据必须被呈现的调度（例如，显示的视频或播放的音频）。 在网络环境中，播放定时约束、特定CM（例如可变比特率（VBR）视频）的速率可变性、网络内分组丢失和延迟的可能性以及有效地使用网络资源的期望一起联合收割机组合使得CM数据的传输成为异常具有挑战性的问题。 这在几个方面与实时连续媒体不同。 首先，CM业务特性通常是提前已知的。 第二，CM数据的特征通常在于关于何时可以开始播放的更宽松的要求。最后，与实时CM数据不同，存储的CM数据不提供响应于网络拥塞而改变编码速率的机会。 在这份提案中，主要研究人员（PI）概述了旨在设计和评估支持存储连续媒体应用的算法和网络传输协议的研究。 这些算法和协议适应并利用所存储的CM的独特特征（例如，已知的业务特性和在其播出时间之前进一步传输数据的能力，CM播出的时间敏感特性）。 PI建议开发和评估提前工作的平滑算法，用于在客户端和网络缓冲限制的情况下将存储的视频传输到客户端。 目标是开发计算CM传输调度的算法，该CM传输调度需要最小的网络资源使用（例如，具有最小的峰值速率要求、最小的带宽要求可变性和最小的缓冲要求）。2)存储的连续媒体的拥塞控制。PI建议开发和评估速率控制算法，该算法自适应地改变存储的CM源的传输速率以响应网络内的拥塞。 特别令人感兴趣的是在干扰CM和非CM连接之间产生的性能权衡。所提出的研究特别强调性能-评估和理解各种设置中的性能，以及正式建立属性（例如，（二）具体方法。 拟议工作中的其他重要主题包括支持客户端交互性的重要性（例如，开发具有小启动延迟的方法，检查启动延迟和资源需求之间的权衡），开发用于各种网络设置的存储CM算法和协议（例如，不同的网络服务模型，以及在单播和多播的情况下），并在所提出的研究中体现的理论和实验的互补组合。

Future high-speed computer networks are projected to carry a dizzying array of networked multimedia applications including digital libraries, video and image servers, distance learning and collaboration, networked virtual environments, and entertainment. A common thread among these applications is the so-called "continuous" nature of their data. In continuous media (CM) data, strict timing relationships exist, which define the schedule by which CM data must be rendered (e.g., a video displayed or audio played out). In a network setting, the playout timing constraints, the rate variability of specific CM such as variable bit rate (VBR) video, the possibility of packet loss and delay within the network, and the desire to use network resources efficiently together combine to make the transport of CM data an exceptionally challenging problem.An important class of continuous media is stored continuous media. This differs from live continuous media in several ways. First, CM traffic characteristics are typically known ahead of time. Second, CM data is often characterized by a more relaxed requirement as to when playout can begin. Last, unlike live CM data, stored CM data does not provide the opportunity for changing the encoding rate in response to network congestion. In this proposal principal investigators (PIs) outline research aimed at the design and evaluation of algorithms and network transport protocols for supporting stored continuous media applications. These algorithms and protocols accommodate and exploit the unique features of stored CM (e.g., known traffic characteristics and the ability to transmit data further in advance of its playout time, the time-sensitive nature of CM playout).The proposed research divides broadly into two areas:1) Smoothing of stored continuous media. The PIs propose to develop and evaluate work-ahead smoothing algorithms for transmitting stored video to a client subject to client and network buffering constraints. The goal is to develop algorithms that compute CM transmission schedules that require a minimal use of network resources (e.g., have the smallest peak rate requirement, the smallest bandwidth requirement variability, and the smallest buffering requirements). 2) Congestion control for stored continuous media. The PIs propose to develop and evaluate rate control algorithms that adaptively vary the transmission rate of a stored CM source in response to congestion within the network. Of particular interest are the resulting performance tradeoffs among interfering CM and non-CM connections.The proposed research has a particular emphasis on performance -- evaluating and understanding performance in a wide variety of settings, as well as formally establishing properties (e.g., optimality) of specific approaches. Other important themes in the proposed work include the importance of supporting client interactivity (e.g., developing approaches with small startup delays, examining the tradeoff between startup delay and resource requirements), the development of stored CM algorithms and protocols for a variety network settings (e.g., different network service models, as well as in unicast and multicast scenarios), and the complementary mix of theory and experimentation embodied in the proposed research.