CAREER: Argus: A Measurement-informed Learning Approach to Managing Multi-cloud Networks

Funding source: NSF CNS-2145813. Period of performance: 06/01/2022 -- 05/31/2027.

Project Overview

Multi-cloud networks are federations of private network infrastructures from the distinct cloud and third-party providers, and serve as increasingly vital underlays for a range of application domains (e.g., genomics, healthcare, high performance computing). Unfortunately, this emerging connectivity paradigm poses significant management barriers to enterprises that seek to deploy overlays and applications due to providers' distinct operational practices, privacy concerns, egress costs, among others. This CAREER project will investigate a novel measurement-informed learning-based framework called Argus to significantly lower the management barriers faced by modern enterprises.

This project will focus on scientific inquiries in three synergistic thrusts to realize the Argus framework. First, it will design calibrated measurement tools and techniques, using which enterprises can gain unprecedented visibility into the federated underlays. Second, adhering to the privacy concerns of providers, it will investigate learning-based modeling capabilities, using which enterprises can accurately infer, localize, and attribute performance bottlenecks to appropriate providers. Third, it will take a principled approach to design a management capability, using which enterprises can effectively and efficiently navigate egress costs and operational goals while avoiding inferred performance bottlenecks.

This project will lower multi-cloud management barriers, enhance the operational productivity of enterprises, and foster breakthroughs in the aforementioned domains and beyond. These will have significant impacts on the economy and society. The research will be tightly integrated with education, emphasizing experiential learning. Activities include inviting underrepresented students from Lane Community College to participate in a mini-research experience, organizing virtual summer schools on project-related topics, involving undergraduates in research, and developing a curriculum on multi-cloud networks. The activities will catalyze community engagement and result in a globally competitive STEM workforce with the necessary skills in this emerging area.

People

  • Sole PI: Ram Durairajan
  • Ph.D. Students: Joseph Colton

Publications

  • ARISE: A Multi-Task Weak Supervision Framework for Network Measurements
    Jared Knofczynski, Ramakrishnan Durairajan and Walter Willinger
    In IEEE JSAC Series on Machine Learning in Communications and Networks, July 2022.
    [PAPER]     [CODE]    

  • ELF: High-Performance In-band Network Measurement
    Joel Sommers and Ramakrishnan Durairajan
    In Proceedings of TMA'21, Virtual, September 2021.
    [PAPER]     [CODE]

  • A First Comparative Characterization of Multi-cloud Connectivity in Today's Internet
    Bahador Yeganeh, Ramakrishnan Durairajan, Reza Rejaie and Walter Willinger
    In Proceedings of PAM'20, Oregon, USA, March 2020.
    [PAPER]    

Outreach