Federated Intelligence for Active Queue Management in Inter-Domain Congestion

Cesar A Gomez,Xianbin Wang,Abdallah Shami

doi:10.1109/access.2021.3050174

Cesar A Gomez, Xianbin Wang + Show 1 more

Open Access

https://doi.org/10.1109/access.2021.3050174

Copy DOI

Journal: IEEE Access	Publication Date: Jan 1, 2021
Citations: 26	License type: CC BY 4.0

Affiliation: Western University

Abstract

Active Queue Management (AQM) has been considered as a paradigm for the complicated network management task of mitigating congestion by controlling buffer of network link queues. However, finding the right parameters for an AQM scheme is very challenging due to the dynamics of the IP networks. In addition, this problem becomes even more complex in inter-domain scenarios where several organizations interconnect each other with the limitation of not sharing raw and private data. As a result, existing AQM schemes have not been widely employed despite their advantages. Therefore, we present a solution that tackles the challenges of tuning the AQM parameters for inter-domain congestion control scenarios where the network management goes beyond an organization’s domain. We then introduce the Federated Intelligence for AQM (FIAQM) architecture, which enhances the existing AQM schemes by leveraging the Federated Learning approach. The proposed FIAQM framework is capable of dynamically adjusting the AQM parameters in a multi-domain setting, which is hard to achieve with the conventional AQM solutions working alone. To this end, FIAQM uses an artificial neural network, trained in a federated manner, to predict beyond-own-domain congestion and an intelligent AQM parameter tuner. The evaluation results show that FIAQM can effectively improve the performance of the inter-domain connections by reducing the congestion on their links while preserving the network data private within each participating domain.

Highlights

Communication over the Internet relies on data packet transmission across a selected network path, while involved over the complex interconnected network elements
We presented our FIAQM solution, which leverages the characteristics of existing Active Queue Management (AQM) schemes in such scenarios where several parties are involved in a communication process and privacy is a major consideration
The main components of the FIAQM architecture effectively applies the fundamentals of the Federated Learning (FL) approach to attain congestion control between Autonomous Systems (ASes) managed by different organizations and whose network data cannot be shared

Summary

INTRODUCTION

Communication over the Internet relies on data packet transmission across a selected network path, while involved over the complex interconnected network elements. A kind of cooperative mechanism is needed to achieve an effective Machine Learning solution where the privacy is paramount: an inter-domain link involves routers at several organizations or geographical regions, which means the possibility of having one or more domains not willing to share their data We propose to place the Learning Orchestrator at the IXP premises, since it is supposed to be a neutral player In this way, FIAQM applies FL to predict the IXP congestion based on the buffer statistics of the intradomain links of the border routers involved (denominated as the Local Learners).

RELATED WORK

FEDERATED CONGESTION PREDICTOR

EXPERIMENTATION DESIGN

FIAQM PERFORMANCE EVALUATION

REAL-TIME AQM TUNING WITH FIAQM

Findings

CONCLUSION