Page 117 - ITU Journal Future and evolving technologies Volume 2 (2021), Issue 4 – AI and machine learning solutions in 5G and future networks

ITU Journal on Future and Evolving Technologies, Volume 2 (2021), Issue 4

ANALYSIS ON ROUTE INFORMATION FAILURE IN IP CORE NETWORKS BY NFV‑BASED TEST
ENVIRONMENT

1
1
1
1
Xia Fei , Aerman Tuerxun , Jiaxing Lu , Ping Du , Akihiro Nakao 1
1 The University of Tokyo, Japan
NOTE: Corresponding author: Ping Du, duping@g.ecc.u‑tokyo.ac.jp

Abstract – Stable and high‑quality Internet connectivity is mandatory for 5G mobile networks. However, the pandemic of
COVID‑19 has forced global and large‑scale staying at home and telecommuting in many countries. The increasing traf ic
has induced more pressure on networks, devices and cloud data centers. It becomes an essential task for network opera‑
tors to enable their ability to automatically and rapidly detect network and device failures. We propose a highly practical
method based on highly practical technology. Our method has a high generalization ability that can ef iciently extract fea‑
tures from large‑scale unstructured data and ensure high accuracy prediction. First, 997 useful features are extracted from
28GB‑per‑day network logs. Then, a differential approach is employed to preprocess the extracted features so as to highlight
the differences between normal and abnormal states. Third, those features are re ined based on the feature importance we
calculated. According to our experiment, the proposed feature extraction and re inement method can reduce computation
without degrading the performance. Among the ive types of failures, we achieve a 100% recall rate in four types and the rest
can also reach 71%. Overall, the total average prediction accuracy of the proposed method is 94%.
Keywords – Core network, failure detection, route information, machine learning

1. INTRODUCTION ternal/external route information and provide appropri‑
ate feedback. Thus, these BGP routers play very impor‑
The pandemic of COVID‑19 has forced global and large‑ tant roles in 5G services, and in order to maintain a cer‑
scale staying at home and telecommuting in many coun‑ tain level of service, it is desirable to immediately detect
tries. The implementation of social restrictions increases hardware and software defects and malfunctions. More‑
Internet traf ic, particularly the traf ic of remote working, over, increasing network traf ic also brings challenges to
web meetings, and online education. For instance, Net lix data‑based optimization. Recent hot AI technologies pro‑
has faced a surge in subscriber numbers, with almost 16 vide novel approaches that are able to migrate our focus
million people signing up for accounts in the irst three of work from fault handling to fault prediction, which al‑
months of 2020 [1]. Zoom’s daily active users spiked lows operators to take precautions in advance.
to 200 million in March 2020, up from 10 million in De‑
cember 2019 [2]. Such increasing network traf ic has in‑ Based on the data sets [4] provided by KDDI Corpora‑
duced more network and device failures than before. For tion, we propose an ef icient method to predict network
example, it is reported that Google has suffered an esti‑ and device failures from large amounts of unstructured
mated $1.7M loss in advertisement revenues during their log iles in real time. Our proposal contains three main
“outage” in December 2020 [3]. Thus, how to automati‑ steps: Feature Extraction, Feature Re inement, and Fea‑
cally and rapidly detect network and device failures has ture Reduction. In the Feature Extraction step, we ex‑
become an essential problem in daily operation. tract 997 features from 28GB per day of unstructured
Network technologies such as 5G, have dramatically log iles, and merge tagged features from the follow‑
ing three kinds of JSON log iles: physical‑infrastructure,
changed the telecommunication environment that brings virtual‑infrastructure, and network‑devices. As for the
faster speed experience to us. 5G mobile networks re‑ BGP‑related entries, we use the number of next‑hops in
quire stable, high‑quality Internet connectivity, but when each array and corresponding pre ixes as features.
a failure happens, the consequences of that failure are ex‑
tremely serious. In addition, since the Internet is oper‑ In order to derive metrics that are changed when there is
ated mutually among ISPs, even if a failure occurs in a an the occurrence of a failure, we highlight the difference
certain ISP domain, the failure spreads rapidly all over between normal and abnormal entries and de ine a new
the world. However, only experienced ISPs can deal with feature named Differential Data to re lect the variations
such a network failure that affects the world. It is de‑ between abnormal data and normal data. After the data
sirable that anomaly detection could be performed au‑ processing, one CSV ile that could be utilized for training
tomatically and promptly. The IP backbone network of or evaluation is generated.
one ISP is interconnected with others via Border Gateway For the sake of importance analysis, the XGBoost [5]
Protocol(BGP) routers. A BGP router needs to continu‑ model is trained by us to calculate the importance scores
ously update the route information from the received in‑ which work as the reference in the Feature Reduction

© International Telecommunication Union, 2021 101