Unde nderstandi nding ng the he Evolvi ving ng Interne rnet - - PowerPoint PPT Presentation

▶
unde nderstandi nding ng the he evolvi ving ng interne
SMART_READER_LITE
LIVE PREVIEW

Unde nderstandi nding ng the he Evolvi ving ng Interne rnet - - PowerPoint PPT Presentation

Apr 23, 2023 472 likes •744 views

Unde nderstandi nding ng the he Evolvi ving ng Interne rnet Ram Durairajan Assistant Professor, Computer and Information Science Co-director, Oregon Networking Research Group University of Oregon 0 In Inter erne net t is is a a co

slide-1
SLIDE 1

Unde nderstandi nding ng the he Evolvi ving ng Interne rnet

Ram Durairajan

Assistant Professor, Computer and Information Science Co-director, Oregon Networking Research Group University of Oregon

slide-2
SLIDE 2

1

Physical Internet

Users, Apps and Data Datacenters and CDNs Cloud Services Mobile Devices SDNs and NFVs Internet of Things

In Inter erne net t is is a a co complex syst ystem

slide-3
SLIDE 3

Ph Physic ical al In Inter ernet

2

Point of Presence (POP) Fiber optics link Datacenter Colocation facility … Submarine cable To London Lit fiber Dark fiber Conduit

slide-4
SLIDE 4

3

Level of Robustness Existing Mechanisms (e.g., traffic engineering) Perfect Connectivity No Connectivity

Pr Problem

  • Gi

Given the claim that In Inter ernet’s d s desi esign i is r s rob

  • bust,

, why do we have

  • u
  • utages? P

es? Per erfor

  • rmance i

e issu ssues? B es? Bandwidth th on

  • n d

dem emand?

(Robustness: ability of the physical Internet to cope with evolution)

slide-5
SLIDE 5

No No o

  • ne h

has a a c complete v view o

  • f t

the I Internet

4

Source: Lumeta Source: Peer1 Router-level Topology Autonomous Systems-level Topology

slide-6
SLIDE 6

5

Level of Robustness Existing Mechanisms (e.g., traffic engineering) Perfect Connectivity No Connectivity

Pr Problem

  • Gi

Given the claim that In Inter ernet’s d s desi esign i is r s rob

  • bust,

, why do we have

  • u
  • utages? P

es? Per erfor

  • rmance i

e issu ssues? B es? Bandwidth th on

  • n d

dem emand?

  • How do we transcend this robustness gap to build a better

Internet?

(Robustness: ability of the physical Internet to cope with evolution)

  • What about evolving components? IoTs? Private interconnects?
slide-7
SLIDE 7

Ou Outline

Introduction and Motivation Unravelling the Structural Complexity

  • Mapping the Internet Ecosystem

Providing Flexible Decision Support

6

slide-8
SLIDE 8

Ma Mapping the Intern rnet ecosystem

  • XConnects, Cloud connects and Private Interconnects
  • Internet of Things
  • Long-haul and Metro

7

slide-9
SLIDE 9

Ma Mapping the Intern rnet ecosystem

  • XConnects, Cloud connects and Private Interconnects
  • Internet of Things
  • Long-haul and Metro

8

slide-10
SLIDE 10

Mapping the Internet of Th Things

  • Map and Characterize the IoT devices and deployments
  • An active measurements-based approach
  • Specific focus on IPv6-enabled IoT devices
  • Challenges
  • IPv6 address space is large. How to efficiently scan IPv6 prefixes?
  • How to differentiate IoT vs. non-IoT devices?
  • Apply this to problems of interest
  • Security and privacy, census and survey, business intelligence, etc.

9

slide-11
SLIDE 11

Ma Mapping long-ha haul ul and nd metro

  • Internet Atlas: a comprehensive repository of the Physical Internet
  • Search-based data
  • Maps nodes, links, fiber strands, etc.
  • Repository has over 1,400 maps
  • Apply this to problems of interest
  • Robustness, performance, security, resilience, etc.
  • Popular Science
  • Best of What’s New, Security Category, 2017
  • One of the 100 Greatest Innovations of 2017

10

slide-12
SLIDE 12

Ma Map of US S long-ha haul ul fiber ber

11

slide-13
SLIDE 13

Asse Assessi ssing ng infr frastruc uctur ure sha sharing ng

  • Striking characteristic of constructed maps is conduit sharing
  • 20-year fiber IRU to reduce costs

12

slide-14
SLIDE 14

Co Connectivity-on

  • nly s

shared r risk

  • How many ISPs share a conduit?

13

100 200 300 400 500 600 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 Raw number Number of ISPs sharing a conduit 542 conduits 89.67% 63.28% 53.50% 12 critical choke points

Physical connectivity lacks much diversity that is a hallmark of commonly-known models.

slide-15
SLIDE 15

Ke Key observation

  • There is a lot of sharing in the Internet
  • Risks and outages
  • Optical connections cannot be reconfigured
  • Inflexibility
  • Risks + outages + inflexibility = NOT robust!

14

slide-16
SLIDE 16

Ou Outline

Introduction and Motivation Unravelling the Structural Complexity Providing Flexible Decision Support

  • Building systems to create a better Internet

15

slide-17
SLIDE 17

Ne Need f for f flexible d deci cision s support

  • Flexible decision support is important
  • Enhance robustness, resilience, security
  • Resilience: remove the inflexible leasing model (and reduce shared risk)
  • Security: connectivity/bandwidth on demand to counter volumetric DDoS attacks
  • Given the understanding of the physical Internet, what radical change

can we introduce to build a better Internet?

  • Wide-area Connectivity as a Service
  • Agility meets the Internet
  • E.g., Deploy NFVs in the wild

16

slide-18
SLIDE 18
  • Objective: a system (called GreyFiber) for cloudification of the

physical Internet

  • Cloud: Rent cycles, use resources, and release
  • GreyFiber: Rent connectivity, transfer data, and release connectivity
  • System considers
  • Infrastructure abundance (e.g., unused fiber)
  • Market economics (e.g., CAPEX, OPEX)
  • Technology trends (e.g., fast remote reconfigurations in routers)
  • Flexible access to fiber-optic paths between endpoints (e.g., IXP) over

a range of use scenarios

17

Wi Wide de-ar area ea Connec ectivity tivity as as a a Ser ervic vice

slide-19
SLIDE 19
  • GreyFiber consists of three components
  • Global control, local site control and physical infrastructure substrate

18

Gr GreyFib iber sy system design

slide-20
SLIDE 20

Fiber Exchange Buyers Sellers Global Controller

GreyFiber Global Control

  • Control and command center
  • Sellers are major fiber/major

cable providers

  • Buyers are the customers (e.g.,

CDNs, enterprise networks)

  • Fiber exchange to enable

economic viability

  • Runs GSP auctions
  • Global controller
  • Traffic engineering
  • Time-based circuit provisioning
  • Network management
  • Backup restoration
slide-21
SLIDE 21

Site A Local Controller Site B Local Controller Site C Local Controller Fiber Exchange Buyers Sellers Global Controller

GreyFiber Global Control GreyFiber Local Site Control

  • Local control over marked

geographic region (e.g., IXP)

  • Mimics minimal functionalities

from global control

  • Configure links
  • Monitor connectivity
  • Report statistics to global

control

slide-22
SLIDE 22

Site A Local Controller Site B Local Controller Site C Local Controller Fiber Exchange Buyers Sellers Global Controller

GreyFiber Global Control GreyFiber Local Site Control Physical Infrastructure

  • Composed of traditional nodes

and links (e.g., fiber paths)

  • Assumption
  • Fiber is already lit
slide-23
SLIDE 23
  • GreyFiber consists of three components
  • Global control, local site control and physical infrastructure substrate
  • Supports a range of use scenarios
  • Small (seconds to minutes), medium (hours), large (days to months) and

extra-large (years)

  • Short lifetime to address unexpected outages and demands
  • Medium-to-large to service unexpected demands without deadlines
  • Extra-large to support traditional lease

22

Gr GreyFib iber sy system design

slide-24
SLIDE 24

Gr GreyFib iber im implem lemen entatio tion an and evalu aluatio tion

  • Implemented in ~22K lines of Python code
  • Evaluated in GENI and CloudLab testbeds

23

slide-25
SLIDE 25

Ke Key results

  • Performance benefits of GreyFiber?

24

1G on GENI 10G on CloudLab

slide-26
SLIDE 26

Qu Questions?

25

Ram Durairajan

ram@cs.uoregon.edu

Thanks to Reza Rejaie, Paul Barford, Joel Sommers, Walter Willinger and “great” students!