可靠组数据传输是数据中心中的一个重要传输模式.在可靠组数据传输问题中,有一个数据源,和一个接收端的集合.可靠组数据传输需要把数据从数据源可靠的分发给所有接收端.P2P是一个典型的可靠组数据传输系统,在互联网得到广泛的应用.然而,P2P系统对可靠组数据传输的完成时间没有保障.与互联网不同的是,数据中心有以下特点:1)数据中心的拓扑结构是预先知道的.2)数据中心可以被集中管理,各个结点之间是相互合作的.基于这两个特点,本文对数据中心的P2P系统进行了重要改进.1)通过一个集中控制算法对每个结点的邻居进行选取.2)在某些特定结点进行数据缓存.我们在ns3平台实现了一个P2P系统,并用本文的方法对其进行改进,在BCube(4,1)中传输4GB的数据,BitTorrent需要52s完成所有数据的传输。而本文的方法只需要18s.
Reliable Group Data Delivery ( RGDD ) is an important traffic pattern in data centers. In the problem of RGDD, there is a data source, and a set of receivers. The problem of RGDD is how to reliably deliver data to all the receivers. P2P is a representative RGDD system, which has been widely used in Internet. However, P2P systems cannot guarantee the finish time of RGDD. Different from Internet, data centers have the following properties: 1 ) The topology is known in advance. 2 ) Data centers can be centralized controlled, and all the nodes are cooperative. Motivated by these properties, this paper proposes important improvements for P2P systems in data centers: 1 ) Uses a centralized algorithm to selected peers for each node. 2 ) Uses packet caching in certain nodes. We have built a P2P system in ns3, and also improved it based on our scheme. We use our system and BitTorrent to transfer 4GB data in a BCube(4,1 ), BitTorrent finishes the transmission in 52s, while our system only uses 18s.