【摘要】：網(wǎng)絡(luò)流量測量是研究網(wǎng)絡(luò)行為的一種有效途徑，是對互聯(lián)網(wǎng)進(jìn)行管理和控制的基礎(chǔ)，在高速網(wǎng)絡(luò)不斷發(fā)展的今天，流量測量面臨著海量數(shù)據(jù)存儲的問題，這就對測量系統(tǒng)的存儲容量和存儲速率提出了極大的挑戰(zhàn)，而基于數(shù)據(jù)流的測量通過將數(shù)據(jù)包按照某種分類原則歸并為流，大大節(jié)省了存儲空間，為流量測量開辟了一個全新的途徑。 研究表明[6]，網(wǎng)絡(luò)中流的總數(shù)雖大，但是流表現(xiàn)出非常強烈的重尾分布特征，即9%的數(shù)據(jù)流占據(jù)了大約90%的字節(jié)流量，因此，了解大流就能很好地掌握網(wǎng)絡(luò)通信的大致信息，近年來，隨著網(wǎng)絡(luò)規(guī)模的不斷增加，網(wǎng)速的空前提高，對大流的研究也顯得日趨重要，大流測量逐漸成為網(wǎng)絡(luò)測量的熱點，研究高效準(zhǔn)確地大流測量算法在當(dāng)下具有非常重要的意義。 本文分對大流檢測相關(guān)算法進(jìn)行了研究，針對哈希技術(shù)、抽樣技術(shù)等關(guān)鍵技術(shù)，對已有算法進(jìn)行了改進(jìn)，融合出一種新的大流檢測算法。具體來說，本文做的研究工作包括： （1）將流量測量分為大流檢測和大流存儲兩個模塊。在大流檢測模塊，對傳統(tǒng)的Counting Bloom Filter（計數(shù)型布魯姆過濾器）進(jìn)行了改進(jìn)，改進(jìn)后的Counting BloomFilter采用了多層結(jié)構(gòu)，相較于傳統(tǒng)的Counting Bloom Filter節(jié)省了大量存儲空間，并能有效解決CBF存在的溢出問題；在大流存儲模塊，使用定長的LRU結(jié)構(gòu)，LRU結(jié)構(gòu)用雙向鏈表來實現(xiàn)，查找效率高，能有效地對檢測出的大流進(jìn)行存儲。經(jīng)理論分析，本文研究的大流測量算法LRU-MCBF占用空間小，時間復(fù)雜度低，并通過仿真實驗驗證了LRU_MCBF在大流測量中漏報率和錯報率較低，能實現(xiàn)高速網(wǎng)絡(luò)環(huán)境下大流對象的準(zhǔn)確提取。 （2）將抽樣算法融合到LRU-MCBF算法中。在高速網(wǎng)絡(luò)的大流檢測中，主機對數(shù)據(jù)包的處理速率的要求是無上限的，實現(xiàn)更高的處理速率是網(wǎng)絡(luò)大流檢測中不斷追求的目標(biāo)，而基于抽樣算法的大流檢測就是一種很好的處理方式，抽樣算法易于實現(xiàn)，并且能在保證一定檢測準(zhǔn)確性的前提下，，大大提高主機處理數(shù)據(jù)包的速率，是一種“低成本，高成效”的測量方式。本文也對抽樣算法同非抽樣算法進(jìn)行了比對，通過實驗證明了其準(zhǔn)確高效的特點，在大流測量中發(fā)揮了非常重要的作用。
[Abstract]:Network traffic measurement is an effective way to study network behavior and is the basis of Internet management and control. With the development of high-speed network, traffic measurement is faced with the problem of massive data storage. This poses a great challenge to the storage capacity and storage rate of the measurement system, and the measurement based on the data flow greatly saves the storage space by merging the data packets into streams according to some classification principle. It opens up a new way for flow measurement. The results show that [6] although the total number of flows in the network is large, the flow shows a very strong heavy-tailed distribution, that is, 9% of the data flow accounts for about 90% of the byte traffic, so knowing the large stream can well grasp the general information of the network communication. In recent years, with the continuous increase of network scale and the unprecedented increase of network speed, the research on large flow becomes increasingly important, and the measurement of large flow has gradually become a hot spot in network measurement. It is of great significance to study the efficient and accurate large flow measurement algorithm. In this paper, the related algorithms of large flow detection are studied. The existing algorithms are improved for the key technologies such as hashing and sampling, and a new algorithm for large flow detection is proposed. Specifically, the research work in this paper includes: (1) the flow measurement is divided into two modules: large flow detection and large stream storage. In the large stream detection module, the traditional Counting Bloom Filter (counting Bloom filter is improved. The improved Counting BloomFilter adopts multi-layer structure, which saves a lot of storage space compared with the traditional Counting Bloom Filter. In the large stream storage module, the fixed length LRU structure is used to realize the bidirectional linked list, and the lookup efficiency is high, and the detected large stream can be stored effectively. The theoretical analysis shows that the large flow measurement algorithm LRU-MCBF has small space and low time complexity. The simulation results show that the LRU_MCBF has a low rate of missing and misreporting in the measurement of large flow. It can realize the accurate extraction of large stream objects in high-speed network environment. (2) the sampling algorithm is integrated into LRU-MCBF algorithm. In the large stream detection of high speed network, there is no upper limit for the processing rate of the data packet in the host computer, and the realization of higher processing rate is the goal of the network large flow detection. The large stream detection based on the sampling algorithm is a good processing method. The sampling algorithm is easy to implement, and can greatly improve the speed of the host processing data packets on the premise of ensuring certain detection accuracy, so it is a kind of "low cost." A highly effective way of measuring. This paper also compares the sampling algorithm with the non-sampling algorithm. It is proved by experiments that the algorithm is accurate and efficient and plays a very important role in large flow measurement.
【學(xué)位授予單位】：江南大學(xué)
【學(xué)位級別】：碩士
【學(xué)位授予年份】：2014
【分類號】：TP393.06

【參考文獻(xiàn)】

相關(guān)期刊論文 前5條

1 裴育杰;王洪波;程時端;;基于兩級LRU機制的大流檢測算法[J];電子學(xué)報;2009年04期

2 王洪波;裴育杰;林宇;程時端;金躍輝;;基于LRU的大流檢測算法[J];電子與信息學(xué)報;2007年10期

3 張連芳,薛飛,舒炎泰;高速網(wǎng)絡(luò)的自相似業(yè)務(wù)模型及其性能評價[J];計算機研究與發(fā)展;1998年06期

4 劉元珍;劉淵;李小航;;基于DCF的資源可控流抽樣[J];計算機工程與應(yīng)用;2008年19期

5 謝鯤;文吉剛;張大方;謝高崗;;布魯姆過濾器查詢算法[J];軟件學(xué)報;2009年01期

本文編號：2203645