Summary
2021
Session Number:TS6
Session:
Number:TS6-1
Detection of Hypergiants in AS-Level Topology Using Machine Learning_
Michiko Harayama, Takuro Kudoh,
pp.178-183
Publication Date:2021/9/8
Online ISSN:2188-5079
Summary:
With the spread of cloud services over the past two decades, the number of content holders (CHs) on the Inter- net has grown. Some of them are huge multinational companies such as GAFAM (Google, Apple, Facebook, Amazon, and Mi- crosoft). In addition, content delivery networks (CDNs) have grown significantly with the increase in traffic from CHs to us- ers. Some of these CHs and CDNs generate traffic levels compa- rable to those of major internet service providers (ISPs) and are called hypergiants (HGs). The impact of HGs is as large as that of major ISPs, and they need to be watched closely because the failure of any one of them may affect the entire Internet. Alt- hough it is not easy to capture the growth of individual autono- mous systems (ASes), the detection of unknown growing HGs will be useful for preventing communication failures and under- standing the impact status of ASes on the Internet.Therefore, in this study we attempted to detect unknown HGs by using publicly available data and machine learning methods. First, we extracted ASes from Tier 1 to Tier 3 from the AS relationship data published by the Center for Applied Inter- net Data Analysis (CAIDA) and analyzed the features of the ASes and the AS-level topology as a complex network. Next, we found that the random forest machine learning method was suit- able for classifying ASes by their features, so we trained the fea- tures of famous HGs and detected HGs by using random forest. As a result, currently growing CDs and CHs were detected as HGs.