Study of Parallel Access Schemes to Speed up the Internet
Study of Parallel Access Schemes to Speed up the Internet
复制标题
加速互联网并行访问方案的研究
DOI：
--
复制
发表时间：
1999
期刊：
影响因子：
--
通讯作者：
Andreas Kirpal
中科院分区：
文献类型：
--
作者：
Andreas Kirpal
互联网的指数级增长使流行的服务器不堪重负，增加了对带宽的需求并增加了客户端的检索时间。为了减少这些问题，多个服务器在地理上分布以存储流行文档的副本。例如，使用网络缓存，原始服务器存储文档的主副本，所有地理上分散的缓存存储和拉取该文档的副本。通过镜像站点复制，文档被复制到辅助站点，以便在服务器之间分配请求负载，并减少客户端的检索延迟。但是，选择最佳服务器并不明显，并且所获得的性能可能会因所选服务器的不同而有很大差异。为了消除复杂的选择过程并减少下载时间，可以使客户端能够从多个源并行下载文档。在这篇论文中，我们研究和评估不同的策略，从多个来源并行下载文档。我们首先考虑具有非常小反馈信息的开环分布。这种情况非常适合无线或卫星通信，其中反馈信道非常弱;它也非常适合多播情况，其中处理来自许多客户端的反馈信息非常复杂。我们研究不同的调度机制，在加速，损失率，源的数量，和源之间的时间位移。为了科普不同来源/客户端的高度异构性，我们还考虑使用前向纠错码。然后，我们开发的分析模型，以评估不同的停止机制在开环循环分布和研究其性能方面的带宽浪费和加速降低。我们考虑一个基于TCP的情况下，客户端服务器协商哪些数据包传输。我们提出了一个新的并行访问方案，自动调度不同的源之间的传输，以尽量减少客户端的检索时间。这种基于TCP的并行访问方案不需要在源处重新编码文档，并且使用现有的网络协议。基于实验结果，我们表明，基于TCP的并行访问方案提供了非常好的性能方面的加速比，消除了需要一个复杂的选择算法和平衡的负载之间的几个来源。
The exponential growth of the Internet is overloading popular servers, increasing the demands for bandwidth and increasing the clients’ retrieval times. In order to reduce these problems, multiple servers are geographically distributed to store copies of popular documents. For example, with network caching, the origin server stores a master copy of the document and all the geographically dispersed caches store and pull copies of this document. With mirror site replication, documents are replicated into secondary sites, in an effort to both distribute the load of requests across servers and to decrease clients’ retrieval latencies. However, choosing the best server is not obvious and the obtained performance can dramatically vary depending on the selected server. To eliminate complex selection processes and reduce the downloading times, clients could be enabled to download a document in parallel from multiple sources. In this thesis we examine and evaluate different strategies to download documents from multiple sources in parallel. We first consider an open loop distribution with very small feedback information. This scenario is well suited for wireless or satellite communications, where feedback channels are very week; it is also well suited for multicast scenarios where it is very complex to handle feedback information from many clients. We study different scheduling mechanisms in terms of speedup, loss rate, number of sources, and temporal displacement between the sources. To cope with the high heterogeneity of different sources/clients, we also consider the use of forward error correction codes. Then we develop analytical models to evaluate different stopping mechanisms in an open loop-loop distribution and study its performance in terms of bandwidth wasted and speedup reduction. We consider a TCP-based scenario where client-servers negotiate which packets to transmit. We present a new parallel access scheme that automatically schedules the transmission between the different sources to minimize clients’ retrieval times. This TCP-based parallel access scheme does not require to re-encode the documents at the sources and uses the existing network protocols. Based on experimental results we show that a TCP based parallel access scheme gives very good performance in terms of speedup, eliminates the need for a complex selection algorithm and balances the load between several sources.
影响因子：
3.7
作者：
Fan, L;Cao, P;Broder, AZ
通讯作者：
Broder, AZ