2017 Technology Exchange

Use Internet2 SiteID

Already have an Internet2 SiteID?
Sign in here.

Internet2 SiteID

Your organization not listed? Create a local account to use Internet2 services.

Create SiteID

Reviewing Machine Learning Approaches used in Wide Area Networks

Time 10/18/17 08:30AM-08:50AM

Room Grand Ballroom A

Session Abstract

Machine learning (ML) algorithms can be used to predict network behavior such as ‘which QoS change will cause what event X with what probability P’. Detecting anomalies in advance can cut down costs and time spent finding misbehaving devices in complex network infrastructures. However, network performance depends heavily on topology configured across devices, paths provisioned, application-network behavior, QoS requirements and underlying traffic restrictions. Making these decisions in near real-time requires massive data processing power and time, such to digest all relevant datasets to recognize multiple outputs that affect the application-network interaction.

Machine learning techniques are not new in network research and have been used to study security, detect malware, DOS attacks, and user-network interaction. We conducted a literature survey of past 10 years to compile a comprehensive list of various methods used across different layers of WAN communications. The results of the survey show algorithms such as k-means, Naïve-Bayes and regression are the most used to perform classification and also prediction in WAN traffic analysis. However, graph optimization algorithms such as shortest path and rule-based decision making are still heavily used for constructing topologies and dedicated links.

In this talk, we present the survey results, along with new research direction for WAN researchers, leveraging SDN and ML for active monitoring, prediction and informed decision-making across distributed network sites. The talk aims to bring out the most pressing challenges that ML can help solve in current WAN infrastructures, such to improve end-to-end connectivity for diverse application and user demand.


Speaker Mariam Kiran ESnet (DOE Office of Science - Energy Sciences Network)

Presentation Media

media item thumbnail Machine Learning in WAN Research

Speaker Mariam Kiran ESnet (DOE Office of Science - Energy Sciences Network)

Primary track Advanced Networking

Secondary tracks Applications for Research

gold Sponsors

bronze Sponsors