Learning Objectives

• Digital Divide
  • Why are we still stuck in the past?
  • What do we need to change the status quo?
  • What is open-access infrastructure?
• AI/ML to fight Digital Divide
  • What are self-driving networks?
  • Why do we need self-driving networks?
  • How can we build them (can we?)?

Digital Divide

Elephant in the Room!

• What is broadband (in USA)?
  

• Around 20 M+ don't have broadband access!

Sadly, digital divide is real!

Cable vs. Fiber

• Cable:
  • DOCSIS 3.1: Upto 10/1 Gbps (shared)
  • DOCSIS 4.0: Upto 10/6 Gbps (shared)
• Fiber:
  • Upto 44 Tbps (shared)

Fiber is better than Cable!

Why Fiber makes sense?

• Physics:
  • Shannon Limit
    • \(C = B.log_{2}(1+SNR)\)
  • Bandwidth –> Frequency:
    • 200-350 THz vs. upto 1.2 GHz
  • Noise:
    • less attenuation and interferences –> High SNR

Fiber is always better!

Why don't we have Fiber everywhere?

• Return on Investments:
  • Fiber ROI is huge but slow (10-20 years)
  • Example: Verizon FiOS required $18B (vs. < $ 10 B) investment for 14 % coverage
• Monopolies (Cable Cabals)
  • Frontier
    • ~13 M customers only had two or less options (EFF report)

This is worrying but not surprising!

What can we do?

• Build Open Access Networks

  • Two-layer:

    • Local government owns and operates the network, ISPs distribute the services

  • Three-layer:

    • an independent party operates the network

Open Access Arrangement

Success Stories

Open Access Support in California (1)

• SB-1130 Telecommunications (Passed on May 26, 2020)
  • Fiber to all:
    • Provide access to high-capacity, future-proof (read fiber) infrastructure to 98 % CA households

Open Access Support in California (2)

• Unserved household areas (scope broadened):

• Scope broadened:

• If 90% households have less than 25/25 (vs. 6/1) Mbps speeds

• Open access:

• deploy infrastructure to support 100/100 Mbps speeds

A step in the right direction!

Challenges

• Getting Service Providers
  • Incentive and expertise mismatch
    • big players have expertise but won't support
• Reputation
  • Lack of expertise –> easy to mismanage
• Price Competition and Consolidation
  • Hard to fight deep pocket services

Government lacks expertise to operate networks!

AI/ML to Fight Digital Divide

Self-Driving Networks

• Key idea:
  • develop networks that can run themselves
  • use AI/ML + programmable switches
• History:
  • gained popularity in recent years
  • Popularized by:
    • Efforts at Google, Facebook, etc.
    • Sonata-like systems and NSF/Princeton workshops

Self-Driving Networks

Original Vision

Closing the control loop!

Technical Push

• Fully, programmable, protocol-independent data planes, and mechanisms to program them (e.g., P4)
• More fine-grained, programmable network measurements (e.g., Sonata)
• Scale-out, distributed streaming capabilities (e.g., Spark, Storm)
• Customizable actions in the forwarding plane (e.g., with P4)

Self-Driving Networks for Open Access

• Operating Open Access Networks
  • Government lacks expertise
• Can we transfer this expertise?
  • Develop AI/ML-based systems that mimic expert network operators

(How )Can we develop self-driving broadband networks?

Self-driving Broadband Networks (Requirements)

• Sense: Scalable and flexible network-measurement infrastructure
• Infer: AI/ML-based solutions to infer diverse networks events in (near) real-time
• Actuate: Automate configuration of different network devices

Self-driving Broadband Networks (Approach)

• Develop programmable interface for active/passive measurements at modems/routers
• Use open-sourced AI/ML tools and edge cloud solutions to analyse network data
• Develop programmable interfaces for dynamically reconfiguring network devices (safely)

Summary

• This class:
  • Digital divide
  • Open access networks
  • Self-driving (broadband) networks
• Next class:
  • Measurements for self-driving networks