In many cases, telecommunications carriers maintain their communication networks by monitoring the normality of equipment, which they assume represents normality of services. Since telecommunications carriers provide a network service to users, they want to directly manage and operate the service itself. However, a service monitoring method cannot be provided on a large scale due to the problems of system size and high cost. In recent years, with the spread of telemetry, individual users' traffic information can be obtained from routers at low cost, which means the service usage status of individual users can be monitored. On the other hand, it is difficult to know whether a service has failed or is simply not being used when no traffic is observed. To overcome this problem, we propose and evaluate a service normality check method by using time series prediction in this paper.

Mobile network failures have occurred many times in recent years. Some network failures become “silent” failures that mobile carriers cannot detect because of incomplete rules concerning failure detection by the network operating system. However, the increasing number of services and devices, and the increasing complexity of the network make it hard to generate rules that cover all network failures. Therefore, monitoring the network performance from a subscriber's perspective is very important. The traditional way to obtain feedback from subscribers is to use call centers and email. However, it is difficult to detect problems early or in their entirety through those channels because subscribers typically do not call a call center until they are certain the problem was caused by a network. In this paper, we discuss a way to monitor a social networking service (SNS) (Twitter in particular) to find out about problems that affect subscribers. A previous study showed the possibility of early detection of network problems by monitoring Twitter. However, since Twitter includes many conversation topics, it is difficult to find tweets that relate to network problems. Searching by a particular keyword is insufficient since it produces a lot of false positive results that contain the keyword but not the topic of the network problem. We solved this problem by using machine learning to suppress the false positive results. We implemented and evaluated a system to detect network failures from Twitter. As a result, we were able to identify 6 out of 6 large network problems and to suppress the number of false positives to only 6 events, whereas keyword matching detected 94 false positive events. Some of the problems were detected faster than through a call center. Furthermore, we conducted research in order to determine the appropriate machine learning algorithm, parameters, and volume of training data. We also propose a method to estimate the location where the tweeters were located.

In this demo, we present a network failure detection system that is constructed by extracting tweets related to network failures and estimating the failure area. The demo is based on the work “Early Network Failure Detection System by Analyzing Twitter Data” [1].

Internet traffic volumes continue to increase rapidly with the spread of the broadband access and the increase in internet backbone capacities. Moreover, the applications provided over the internet have become more diverse. Adequately supporting the QoS requirements for these diverse applications requires measuring traffic volumes by application type(file transfer, broadcast and so on). Therefore, methods for classifying applications become extremely important for multi QoS environment. In some application traffic classification methods, the applications are analyzed on the basis of statistics for the traffic characteristics, such as the packet inter arrival time and packet size. In this paper, we propose application traffic classification method applying LPC cepstrum. The proposed method enables to treat the traffic data as a time series signal and to separate spectral envelope information and spectral fine structure information from spectrum. We show that it is effective to use the LPC cepstrum to classify internet traffic by comparing LPC cepstrum with flow and raw data in this paper.

With the wide spread of mobile devices, information discovery methods over mobile ad hoc networks (MANETs) are required. Recent studies have revealed that integrating a distributed hash table (DHT) substrate and network-layer routing can increase search efficiency. Furthermore, some researchers have proposed a clustering method which groups the overlay nodes according to their physical distance. However, it has also been pointed out that the search efficiency deteriorates in highly dynamic environments. In this paper, we propose a method sharing pointers among nodes in a cluster. Through simulation experiments, we show that the proposed method improves the success ratio of object search up to 40 % compared with existing method.