Summary
2021
Session Number:PS1
Session:
Number:PS1-7
BRAS Syslog Pattern Generation Method: A Preliminary Experiment on Clustering Algorithms
Yun-Jie Li, Jhao-Yin Li,
pp.25-28
Publication Date:2021/9/8
Online ISSN:2188-5079
Summary:
Internet service provider (ISP) uses Broadband remote access server (BRAS) to connect its customers called subscribers. Due to the usage of the Internet had grown up rapidly, the ISP company have to pay more attention to manage BRAS device. In general, the ISP administrators set up several suspicious Syslog extraction patterns into the operation support system (OSS) to extract Syslogs. The administrators will check out the detail of the Syslog matched the suspicious Syslog extraction pattern to find out if any problem exists on devices while they are notified by OSS that there are some words in Syslog are matching the extraction pattern.However, it is difficult to define the proper extract pattern that the administrators do not see before. Especially, after configuration had been changed, e.g. after an upgraded new version of the software or after adjusted the topology of the network, etc, there are a few new words related anomaly in Syslogs and those words are not match the Syslog extraction pattern because the administrators do not see those before.
Furthermore, we found that BRAS Syslog data have a special feature that the other log data would not have is that there is a lot of content-related digits. Note that, those contents carry unique information and can make the extraction algorithm misleading. According to that, how to leverage the automatically extract algorithm to deal with this problem is another issue we have to take care of.
In this paper, we proposed the BRAS Syslog pattern generation methodology and conduct a preliminary experiment to evaluate the performance. The result shows that the method with the combination of tfidf and EM or tfidf and HCA can produce better performance compared with the other clustering algorithms.