【摘要】：數(shù)據(jù)中心作為云計(jì)算的核心支撐部分，數(shù)據(jù)中心的性能直接決定了云計(jì)算服務(wù)的質(zhì)量。數(shù)據(jù)中心網(wǎng)絡(luò)作為數(shù)據(jù)中心中的通信橋梁，其負(fù)載正變得越來越重，并開始影響數(shù)據(jù)中心的性能。組播在數(shù)據(jù)中心中的擴(kuò)展受到了越來越多人的關(guān)注。組播可以有效的降低數(shù)據(jù)中心中的網(wǎng)絡(luò)負(fù)載，提高網(wǎng)絡(luò)的性能。本文主要研究了數(shù)據(jù)中心網(wǎng)絡(luò)中的組播路由優(yōu)化問題和可靠性組播中的恢復(fù)節(jié)點(diǎn)和路徑選取的問題。 數(shù)據(jù)中心網(wǎng)絡(luò)的架構(gòu)方式主要有兩種，一種是以交換機(jī)為核心，另一種以服務(wù)器為核心。其中以服務(wù)器為核心的架構(gòu)中服務(wù)器也參與數(shù)據(jù)包的轉(zhuǎn)發(fā)，，如果服務(wù)器的負(fù)載壓力很大，這時(shí)對(duì)數(shù)據(jù)包的接受和轉(zhuǎn)發(fā)都會(huì)造成影響，比如增大數(shù)據(jù)包的發(fā)送時(shí)延，增大丟包率等。所以在生成組播樹的時(shí)候，如果考慮到服務(wù)器的負(fù)載壓力的狀況，不僅可以使組播樹更加健壯，同時(shí)也起到了調(diào)節(jié)數(shù)據(jù)中心服務(wù)器之間負(fù)載均衡的作用。本文提出了一種基于服務(wù)器節(jié)點(diǎn)負(fù)載和距離作為優(yōu)化目標(biāo)的新的組播樹生成算法。該算法使用蟻群算法和遺傳算法相結(jié)合的算法，將服務(wù)器負(fù)載和距離當(dāng)做衡量適應(yīng)值的參數(shù)，找出最符合條件的解作為有效結(jié)果。該算法盡量避免組播樹的邊經(jīng)過負(fù)載過大的服務(wù)器節(jié)點(diǎn)，有效地降低了組播的時(shí)延并減小丟包的概率。 傳統(tǒng)可靠性組播的數(shù)據(jù)包恢復(fù)方法是通過組播樹的原有路徑來恢復(fù)。但針對(duì)數(shù)據(jù)中心網(wǎng)絡(luò)中擁有豐富冗余鏈路的特點(diǎn)，現(xiàn)在提出的數(shù)據(jù)包恢復(fù)的方法都是一對(duì)一單播的方法。這樣可以有效的避免通過組播樹中原路徑來進(jìn)行數(shù)據(jù)包的恢復(fù)，不僅減小了原路徑的負(fù)擔(dān)，還增大了數(shù)據(jù)包恢復(fù)的成功率。本文提出了一種基于丟包節(jié)點(diǎn)和恢復(fù)節(jié)點(diǎn)位置優(yōu)先級(jí)的一種恢復(fù)節(jié)點(diǎn)的選取算法。通過節(jié)點(diǎn)本身的位置確定該節(jié)點(diǎn)容易丟包的程度，讓丟包概率越大的節(jié)點(diǎn)優(yōu)先選擇最佳的恢復(fù)節(jié)點(diǎn)，建立節(jié)點(diǎn)之間的一對(duì)一恢復(fù)關(guān)系。該算法增加了易丟包節(jié)點(diǎn)的數(shù)據(jù)包恢復(fù)能力，使各節(jié)點(diǎn)間的恢復(fù)壓力平均，并且占用更少的鏈路資源。 最后通過進(jìn)行仿真實(shí)驗(yàn)，驗(yàn)證了兩個(gè)算法的正確性和有效性。
[Abstract]:Data center as the core support part of cloud computing, the performance of data center directly determines the quality of cloud computing services. As the communication bridge in the data center, the load of the data center network is becoming heavier and heavier, and the performance of the data center is beginning to be affected. More and more people pay attention to the expansion of multicast in data center. Multicast can effectively reduce the network load in the data center and improve the performance of the network. This paper mainly studies multicast routing optimization in data center network and recovery node and path selection in reliable multicast. There are two main architectures of data center network, one is based on switch and the other is server. In the server-centered architecture, the server is also involved in the forwarding of packets. If the load of the server is very heavy, then the acceptance and forwarding of packets will be affected, such as increasing the transmission delay of packets. Increase the packet loss rate, etc. So when the multicast tree is generated, if the load pressure of the server is taken into account, not only can the multicast tree be more robust, but also it can adjust the load balance among the servers in the data center. In this paper, a new multicast tree generation algorithm based on server node load and distance is proposed. The algorithm uses ant colony algorithm and genetic algorithm, takes the server load and distance as the parameters to measure the fitness value, and finds out the most qualified solution as the effective result. The algorithm tries to avoid the edge of the multicast tree from passing through the server node with too much load, which effectively reduces the delay and the probability of packet loss. The traditional packet recovery method of reliable multicast is to recover through the original path of multicast tree. However, in view of the rich redundant links in the data center network, the proposed packet recovery methods are one-to-one unicast methods. In this way, the packet recovery through the central path of the multicast tree can be effectively avoided, which not only reduces the burden of the original path, but also increases the success rate of packet recovery. In this paper, a recovery node selection algorithm based on packet loss node and recovery node location priority is proposed. The degree of easy packet loss of the node is determined by the position of the node itself, and the node with the higher probability of packet loss is allowed to select the best recovery node first, and the one-to-one recovery relationship between the nodes is established. The algorithm increases the packet recovery ability of packet-prone nodes, makes the recovery pressure between each node average, and occupies less link resources. Finally, the correctness and effectiveness of the two algorithms are verified by simulation experiments.
【學(xué)位授予單位】：哈爾濱工業(yè)大學(xué)
【學(xué)位級(jí)別】：碩士
【學(xué)位授予年份】：2013
【分類號(hào)】：TP308

【參考文獻(xiàn)】

相關(guān)期刊論文 前1條

1 丁建立,陳增強(qiáng),袁著祉;遺傳算法與螞蟻算法的融合[J];計(jì)算機(jī)研究與發(fā)展;2003年09期

本文編號(hào)：2472847