Riding along with the Time-Traveling Networking Researcher

Mostafa H. AmmarSchool of Computer Science, Georgia Tech

ammar@cc.gatech.edu

Ø 1985 Ø2018

The Time Traveling Networking Researcher

Ø 1985 Ø2019

The Time Travelling Networking Researcher

Then and NowContent Distribution

1985 2019

Then and NowHow we did research

ØACM SIGCOMM 1985

Ø22 papersØAuthors/paper = 1.9Min =1, Max = 3ØSample Title:“Window Selection in

Flow Controlled Networks”

ØACM SIGCOMM 2019

Ø32 papersØAuthors/paper =6Min = 1, Max = 17ØSample Title:“Gentle flow control:

avoiding deadlock in lossless networks”

Then and NowResearch Approach

ØDesign, Build and Enjoy

ØBuild, Measure,Tweak, Repeat Forever

Then and NowMy Research

What did the Journey look like?

Caveats

ØAdding structure and not re-writing history.

ØHindsight is 20/20ØIt is difficult to make predictions,

especially about the future!

A Story in Three ChaptersØHow to say “Necessity is the Mother of Invention”in Networkingese?

Ø From many, one

Ø From one, many

Takeaways

ØA fundamental iterative process governs successful evolution in deployed infrastructure.

ØThere is nothing fundamental about a single global network delivering all services.

ØNew ManyNets world is upon us:§ flexibility bodes well for the future of networking

and networking research.

“Necessity is the mother of invention”

Ø In Network Speak

§ Necessity == Service Support or Scale

§ Invention == Infrastructure Deployment

The Service-Infrastructure Cycle

New Network Service orIncreased Scale

Establish/Upgrade Infrastructure To Meet Scale or Provide Service

Motivators

ØMesh, any-to-any connectivity

ØScalability

The Service-Infrastructure Cycle

ØWorked for almost 40 years to produce the current Internet

ØUntil in 2005 …complaints about “the inability [of the Internet] architecture to adapt to new pressures and requirements."

Ossification

Ossification Explained

New Network Service orIncreased Scale

Establish/Upgrade Infrastructure To Meet Scale or Provide Service

Ossification is when cycle gets stuck

But the Cycle is fundamental

ØCase Studies§ Unicast Routing§ Multicast Routing

The Cycle in Action: Unicast Routing

ØUnicast Routing: How to find and deploy network paths to single destination.

ØAs the Internet grew routing became more complicated

Internet Growth

Prosser D. V. Link State EGP BGP

Ossification

Cycle in evidence

ØBGP was “built on experience gained with EGP as defined in RFC 904 and EGP usage in the NSFNET Backbone as described in RFC 1092 and RFC 1093 "

The Case of Multicast

ØMulticast: is the act of sending a message to multiple receivers using a single local “transmit” operation.

ØIt is highly Scalable

Unicast and Multicast

Data Copying

Multicast Saves

Multicast and Content Distribution

ØDevelopment anticipated the cycle and did not follow it.

ØNot widely deployed today

Multicast anticipated the Cycle

Aguilar Large Groups

Ossification

PIM SSM

Dalal &Metcalfe

So …

ØBefore Ossification, the Cycle operated to continuously enable novel applications and increased scale.

ØSuccess when Cycle is followed

ØCycle is inoperable due to ossification

From Many, One

A single global infrastructure consumes all services

Then: Many Networks

Ø ca. 1987Ø Tymnet (USA –public)Ø GTE Telenet (USA- public)Ø Transpac (France - public)Ø Datapac (Canada – public)Ø ARPANet (USA – research)Ø BNA (Boroughs)Ø SNA (IBM)Ø DNA (Digital)

ØA network for workØA network for home(AOL)ØA network for

experimentation

The One Network Revolution

ØHaving one network for everything is not an obvious choice.

ØUnder-appreciated networking milestone

ØWas driven by§ Connectivity as the main goal § Need for Scalability§ Economies of scale

OneNet effect on networking research

ØStifled Innovation: Ideas rejected if§ Not deployable on Internet § Not Scalable

ØDifficult to validate without disrupting infrastructure§ Eternal quest for experimentation

environments§ (MBone, QBone, 6Bone, PlanetLab, GENI)

OneNet Contributed to Ossification

ØBy Definition: ManyNets cannot be ossified

So …

If we have OneNet and it is ossified,are we done yet?

From One, Many

How the OneNet is fragmenting

The Service-Infrastructure Cycle

New Network Service orIncreased Scale

Establish/Upgrade Infrastructure To Meet Scale or Provide Service

The Unstoppable Force

ØThe demands on the network continue to evolve§ Increased scale§ Content Providers reign supreme§ Application-specific requirements, low

latency, prioritization

What Happens When …

Unstoppable Force

Ossified Immovable Object

The Fragmentation of the Internet

ØThe Ossified Internet is yielding

ØAre we seeing the return of ManyNets?§ Let’s look at the early signs

Is Google using the Internet?

The Content-Delivery Network

ØShift from Connectivity to ContentØContent Provider

§ Needs to maximize quality of experience§ Reduce reliance on intermediaries§ Control interaction

The Flattening of the Internet

41

End Users

Access ISPRest of the Internet

CP Extensions in Access Networks

ØEmbedded Content Provider Appliances

ØISP Interfaces to allow CP Coordination and Collaboration.

ØA large majority of traffic sees a tiny portion of the global Internet.

“Zero-Hop” Networks

End Users

Access ISPRest of the Internet

Dedicated (Bypass) networks

Ø SpreadNetworks

Ø SIGFOX

Ø Haste

Ø FirstNet

ØAlthea

5G SlicingØ5G = Future Mobile Broadband and

Beyond§ Enhanced mobile broadband (eMBB)§ Ultra-reliable and low latency

communications (URLLC) § Massive machine type communications

(mMTC)

http://www.huawei.com/minisite/hwmbbf16/insights/5G-Nework-Architecture-Whitepaper-en.pdf

What does the ManyNets world look like?

ZeNISP

Bypass

CommunityISP

Users

Servers

ISP

Rest of the Internet

CP Bypass

So What?

ØFragmentation is a reality let’s embrace it.§ Attempts to fight ossification without

fragmentation failed (see “Clean Slate”)§ Fragmentation opens up the research

agenda.ØLoss of OneNet advantages:

§ Economy of scale§ Low barrier to entry

Takeaways

ØA fundamental iterative process governs successful evolution in deployed infrastructure.

ØThere is nothing fundamental about a single global network delivering all services.

ØNew ManyNets world is upon us:§ flexibility bodes well for the future for

networking research

Epilogue:The Exciting Future of Networking

ØScale and new services will continue to drive evolution

ØOssification will be a thing of the past§ The ManyNets world will be a fertile

ground for innovation§ Network Programmability will add to its

responsiveness

Onwards to 2050

The Internet License Office

Internet Licenses

ØStarted being required in 2035!

ØJurisdictions came to the realization that un-regulated highly scalable digital connectivity can be harmful.

ØSolution: Scope Fragmentation and Speed Control.

Internet License Categories

ØA: Close family membersØB: Digitally Gated CommunityØC: Access to non-curated spaces(Chain of digital custody certification,Not Blocked By Default)ØD: Fast Media certificationØE: Access to Space/Time Browser

Proceedings

CoNext2050

Nuuk, Greenland

Thank You!