Research On Business System Discovery Method In Server Communication Network

With the development of the Internet Technology and the rising of the concept of Big Data, Internet Data Center (IDC, data center) market is facing a huge change from the traditional, independent, decentralization to large-scale, centralized. In this age of information explosion, the demand for data storage and the requirements of data processing are rapidly expanding. The capabilities of large data storage and processing will be the key competitiveness of enterprises in the future. Internet Data Center is becoming a key element to a healthy and stable development of the Internet. And it is to provide the basis devices for the efficient Internet server. However, faced with an increasingly scale of data center server clusters, the difficulty of its operation and maintenance is growing faster. Traditional data center operation and maintenance are in accordance with the standard ITIL processes. The processes are complexity and long cycle, and often resulting in the revision of the reach of change information. Because the logical relationship between servers are dependent on the information registered by user. Therefore, the logic of the business situation will lead to chaos. This old method has been unable to adapt to the operation and maintenance of large-scale data centers. So, it needs a highly efficient automated method to sort out the business logic structure of the data center servers.This paper uses the concepts of complex networks to analysis large-scale data centers, not just focus on a single server’s running state, but also concern about the interactions between servers. Communication behaviors between servers are basically made by business actions. The communication behaviors sometimes will reflect their business relationships. And business systems appear as clusters, there will be some network structure. Therefore, the analysis method of complex networks can be used to try to solve the data center business systems’finding problem. First, build an interactive communication network for the data center, and reducing the data center business practices. Then, by analyzing the interaction patterns between servers, building the relationship modular between the servers that used to measure the strength of the relationship. We use weights to reflect the strength of the relationships. And the weights are learning from the data set. Finally, we use community detection algorithm to divide this communications network into parts.We did the experiments on a real data center’s log data that collecting from servers running state. Finally, the experimental results show that the method we designed is feasible. We designed multi-group comparative experiments:the weighted network and the non-weighted network, weights learning methods based on logistic regression and the method based on port features. By comparing the different modeling methods and different community detection algorithms, it can fully demonstrate that the relationship modeling scheme we make can improve the accuracy of the community detection algorithms.