Collaborative Research: Visualization: Overlay Network Support for Remote Visualization of Time-Varying Data

在这个项目中，我们解决了为时变数据的远程可视化提供系统支持的相关问题。大多数现有的远程可视化系统都采用了客户机-服务器模型。由于集中的数据访问和可视化计算，服务器可能成为瓶颈。此外，由于需要传输大量的可视化数据，网络很容易过载。为了克服这些问题，研究人员提出了各种方法，通过将工作负载转移到客户端来分散远程可视化计算。但是，这种工作负载迁移的效果是有限的，因为客户机的计算能力经常受到限制。减少服务器瓶颈的另一种替代方法是在网络中部署多个服务器。然而，部署和管理多台服务器的成本非常高，并且可能引发一些策略问题。我们的目标是开发一个可伸缩的、轻量级的、易于部署的体系结构，它能够为时变数据的远程可视化提供一定程度的服务质量（QoS）。为了实现这一目标，我们提出了一种基于点对点（P2P）系统的架构。一般来说，P2P系统中的每个节点都被认为是平等的，因此该模型在技术上不同于客户机-服务器模型。我们的体系结构的亮点是：(1)服务器数据和工作负载分布到Internet上愿意共享其资源的志愿节点组。这些节点组成多个虚拟服务器。(2)允许客户端通过任播服务访问服务器组。Anycast自动查找组中的最佳服务器，并将数据返回给请求客户机。最好的服务器应该提供最好的服务质量（即数据访问延迟最小）。(3)采用一种时空多分辨率数据结构，即时空分区树（Time-Space Partitioning， TSP），其中存储多个分辨率块，子树上的数据相互独立。通过在树中存储多个分辨率块，客户端在选择最佳服务器时可以有更多的选择，这可能会随着系统动态而变化。在本项目中，我们计划全面设计、分析、实现和评估基于p2p的时变数据远程可视化架构。我们将开展以下研究任务：(1)覆盖网络中qos感知任播服务的设计、分析和实现；(2)改进了TSP树结构，特别是其数据编码方案；(3)设计、评估和实现一种有效的时变数据在志愿节点间的分布算法；(4)基于所提出的体系结构构建和评估原型系统。

In this project, we address issues related to providing system support for remote visualization of time-varying data. A majority of the existing remote visualization systems have adopted the client-server model. The server may become a bottleneck due to the centralized data access and visualization computation. In addition, the network can be easily overloaded because of the large volume of visualization data to be transmitted. To overcome these problems, researchers have proposed various methods to decentralize the remote visualization computation by shifting workload to the clients. The effect of such a workload migration however, is limited since the client 92s computation power is often constrained. Another alternative to reduce the server bottleneck is to deploy multiple servers in the network. However, deployment and management of multiple servers is very expensive, and may raise some policy issues.Our objective is to develop a scalable, lightweight, and easily deployable architecture, which is capable of providing a certain degree of quality of services (QoS) to remote visualization of time-varying data. To achieve this objective, we propose a Peer-to-peer (P2P) system-based architecture. Generally speaking, each node in a P2P system is considered equal, and hence this model is technically different from the client-server one. The highlights of our architecture are (1) Server data and workload are distributed to the groups of volunteer nodes in the Internet which are willing to share their resources. These nodes compose multiple virtual servers. (2) The client is allowed access to the server group via anycast service. Anycast automatically finds the best server in the group, and returns the data to the requesting client. The best server is supposed to provide the best quality of services (i.e. in terms of the minimum delay in data access). (3) A spatio temporal multi-resolution data structure, called Time-Space Partitioning (TSP) tree, where multiple resolution blocks are stored and the data on the sub-trees are independent of each other, is employed. By storing multiple resolution blocks in the tree, a client is permitted more choices in making the selection of best server, which may vary with the system dynamics.In this project, we plan to fully design, analyze, implement, and evaluate the P2P-basedarchitecture for remote visualization of time-varying data. We will carry out the following research tasks (1) design, analysis, and implementation of QoS-aware anycast services in the overlay networks; (2) enhancement of the TSP tree structure, especially its data encoding schemes; (3) design, evaluation, and realization of an efficient algorithm for distribution of time-varying data among the volunteer nodes; (4) construction and evaluation of a prototype system based on the proposed architecture.