| Age of twenty first century is an age of information. With the development of technique of information, network and computer science.More and more people have got to know network. With network, work,study and life become more convenient to us than before. As the development of network, the bandwidth of network has also gained an explosive growth. It is positive, but it is also a challenge for software under high-speed network.The first thing that has to be taken into consideration is the collecting technique of network traffic. In order to analyze the status of the network, we need to use traffic collecting technique. The technique is widely used in traffic monitoring and Network security. However, as the growth of the bandwidth, traditional technique of traffic collecting begins to expose more and more defects. In order to improve this kind of situation, more and more scholars and professors across the world devote themselves to the new technique of traffic collecting. Thankfully,technique of DPDK (Intel Data Plane Development Kit)make it possible to collect the traffic under network of high-speed with high performance.In this thesis, you can see a brief introduction of DPDK.Among all kinds of data analyzing techniques, hashing scheme has taken an important place. Traditional hashing scheme is of high-efficiency in data storing. However, under specific situation such as high-speed network of high uncertainty, traditional hashing scheme lacks in query performance and robustness. In the thesis, we aim to optimize the performance of hashing schemes under high-speed network. We are going to optimize the hashing schemes from three aspects. Firstly, we propose two kinds of cache schemes: PBS-CPE and Cuckoo_Cache. This two kinds of schemes are able to improve the query performance of hashing schemes without change the distribution of elements. Secondly,in order to improve the insert performance of Cuckoo hashing scheme,we propose two ways to solve it: increasing the number of hashing functions and bringing in the overflow stack. At last, we propose two schemes to improve the performance of segmented hash: scheme of the smallest index and iterative updating of Bloom filter. And with the two schemes, we propose a hashing scheme called segmented hashing scheme with skew property. The scheme is of better performance than segmented hash under high speed network.In this thesis, we designed a system of network traffic managing.The system uses technique of DPDK to capture traffic and technique of hashing schemes to store packets. In the meantime, the system can generate statistic of various kinds of hashing schemes. In the end of the process of traffic capture, the statistic is written into the log.We will evaluate the performance of different kinds of optimization design and different hashing schemes in Chapter Six. From the statistic generated by the system, we can find out that PBS-CPE and Cuckoo_Cache can improve the performance of query operation of hashing schemes, optimization of the Cuckoo hashing scheme can greatly improve the insert performance of it, the iterative updating scheme of segmented hash can diminish the false-positive rate of Bloom filters so that the query performance is improved. Segmented hash with skew property is of better performance than segmented hash under specific situation of high-speed network. |
