10 fn key1

The Future of the Internet

Johna Till Johnson
President & Sr. Founding Partner
Nemertes Research
johna@nemertes.com

Agenda

 Introductions
 A Short Personal History of the Internet
 The Internet: Current (High-Level) State
 Internet Challenges
 The Future
 Summary and Takeaways
 Q&A

2 © Nemertes Research 2010 www.nemertes.com 1-888-241-2685 DN1063

Introductions: About Nemertes
 Analyze business value of
emerging technologies
 Benchmark reality
 Advise and assist enterprises,
service providers on:
 Architectures, cost models, business &
organizational strategies

 Core topics:
 UC, Collaboration, Social Computing
 WAN/LAN & Wireless
 Managed & Hosted Services
 Application Delivery Optimization
 Cloud, Virtualization, Data Center
 Security & Compliance


Introductions: About Me
 Engineer, physicist
 Former CTO
 Current president & CEO
 Serve as trusted advisor to
Fortune 500 companies &
others since 1995
 Failed science fiction writer
 Successful kayaker


A Short (Personal) History of the Internet
 1986: DECnet for engineering projects
 Early chat session
 1987: ARPAnet for particle physics projects
 Data transfer between Fermilab and University of Rochester
 1988-1990: Working in industry (no Internet)
 1990-2000: Researching Internet,
 1992 award for next-generation Internet article
 Participating in/with IETF—IPv6, VPNs, MPLS
 2000-2002: Building Internet (sorta)
 MCI, Verizon, Gx backbones
 2002—present: Researching business use of Internet


The Internet

Used with permission: Hebrew University


The Internet Is A Stunning Success!

 Economy increasingly Internet-based
 300% increase in Internet users from 2000-2008
 By all metrics, traffic on the Internet continues to grow
dramatically
 Andy Odlyzsko, MINTS: 50-60% year over year
 AT&T: 65% year over year
 Nemertes: 100% year over year
 Cisco: 200% year over year
 New applications emerge daily
 YouTube, Facebook,Twitter, etc etc etc
 Performance is generally pretty good


Everything’s Great, Right?

Not quite. The Internet is facing
three distinct (though related)
challenges:
1. Exponential growth in
demand
2. Access bandwidth constraints
3. Architectural weaknesses


Challenge 1: Demand Growth


Challenge 1: Demand

 In 2007 and again in 2008, Nemertes modeled
Internet capacity and demand
 We were the only organization to:
 Look at both demand and capacity independently
 Validate models against actual data (go figure!)
 Our findings? Demand will outstrip capacity in
approximately 2011


Internet Infrastructure Models 2007-2008

Our goals:
 To arrive at a defensible estimate of current and
projected Internet traffic:
 Independent estimation of supply and demand
 Model with adjustable knobs to support new data
 To compare with current and projected Internet
infrastructure and investment.
 To achieve order of magnitude accuracy.
 To extrapolate capacity and demand numbers into
investment numbers.


The High-Level Results: What We Found


What We Did NOT Find


Unique Components of Our Model

 “Demand” is essentially “user appetite for bandwidth”,
quantified as utilization of Internet-attached bandwidth
 Includes both “home” and “work” users (distinction is increasingly blurred)
 “Demand” is NOT:
 Measured traffic on the core (Odlyzko)
 Projected traffic generated by applications currently in use (Cisco)
 Why?
 Traffic that makes it to the core could have been gated at the edges
 Projecting traffic based on applications in use today ignores potential for
rapid development of applications in future (eg YouTube)


Demand Validation and Verification


Why Is Demand Exponential?

Multiple growth curves:
 More people (increasing population). Linear, steeper slope outside
US…
 Greater percentage of all people are Internet-connected. Linear…
 Internet-connected people are increasingly connected by more
than one device (see upcoming slides)…
 Bandwidth of Internet-connected devices increasing
exponentially…
 Measured usage of Internet-connected devices increases
exponentially….
…(More internet-connected people) + (more Internet-connected
devices per person) + (increased bandwidth of Internet-connected
devices) + (increased utilization of bandwidth of Internet-
connected devices) = exponential growth!


Internet-Capable People and Devices


The 2005 Inflection Point

Year 2000 2005 2010 2012

Device-to- 0.67 1.53 2.26 2.68
user ratio

A “device” includes: Inversion occurs in approx 2005
 Home PC
 Work PC
 Smartphone/mobile device
 IPTV player
 Internet-attached gaming console

© Nemertes Research 2010 www.nemertes.com 1-888-241-2685 DN1063
19

Conclusions: Demand

 Our model predicts that demand will outstrip capacity,
particularly access capacity, within the next 3-5 years.
 Our model agrees with the best available measured
data to date.
 Our model does not rely on the development of any
one particular application; rather, we assume that
bandwidth-intensive applications will develop
spontaneously and grow exponentially (e.g. YouTube).
 We are NOT predicting that the Internet will break in
2011! We believe that users will find current access
capability insufficient to run the current application
portfolio (imagine running YouTube over dial-up).


Challenge 2: Access Capacity Constraints


Challenge 2: Capacity

 The key reason demand outstrips capacity is lack of
access capacity
 This is due to the cost of investing in “last mile”
technologies (including fiber, copper, and wireless)
 Recent regulatory actions will increase the cost and
decrease the likelihood of investment (FCC claiming
Title 2 regulatory powers)
 Capacity issues likely to get worse, not better!


Challenge 3: Architectural Limitations


Challenge 3: Architecture

 The basic architecture of the Internet has reached its limit.
Scaling further (to accommodate even the most modest
growth projections) will be:
 Technically difficult
 Increasingly expensive
 Ultimately futile
 Technically difficult: Increasing amounts of time and
engineering cycles will be required simply to deliver the
same services and applications available today
 Increasingly expensive: All that engineering takes money
(with no corresponding revenue increase to justify the cost)
 Ultimately futile: Despite ever-increasing investments of
time, money, and engineering horsepower, the Internet will
stop working


Why Does All This Happen?

 Two things are happening at the same time:
1. We’re running out of IP addresses. Because the current design more-or-less requires
globally-unique addresses for every Internet connection, to handle ongoing growth
more addresses are required
2. Routing table size is increasing faster than routing capacity. The bigger the Net gets,
the more routing computation is required to get packets where they’re going
 Solving one problem just exacerbates the other (push on one
side of the balloon and it bulges on the other)
 Making the address space bigger pushes route-table growth past routing capacity
 Limiting routing-table growth means limiting the number of addressable connections
 There’s no way to get globally-unique addresses for every
connection and reasonably bound route-table growth


How the Net Breaks

 Fragmentation: You can’t get
there from here
 Performance degredation:
Voice, video across 3-second
latency? Good luck with that!
 Functional incapacity:
Multihoming? Sorry, you’re
SOL.
 Security: Already gone. An
unprotected host is now
“owned” within 30 seconds.
And considerably more
money is pouring into
organized attacks than the
defense. The barbarians are
overtaking the city
 Reliability: Roving brownouts.


But None of That Is Happening!

It will.


Sez Who?

 Vince Fuller, Cisco
Systems (2006): “Current
trends in the growth of
routing and addressing
state in the global Internet
may not be scalable in
the long term.”
 John Day, Boston
College (February 2009):
“We are now confronting
the most severe threat to
the Internet to date.”


But IPv6 and Moore’s Law Will Save Us!

 IPv6: 128-bit address space (versus 32 for IPv4).
Enough to address five million billion times the number
of visible stars (why this is an important metric is
beyond me, but okay). Purportedly allows the Internet
to scale to far more endpoints than today.
 Moore’s Law: The more routes there are, the bigger
routing tables have to be. But fortunately, no matter
how fast routing tables increase, processing power and
memory capacity of routers increases faster. Right?


IPv6 Fixes Everything! Doesn’t It?

 No.
 Not only is IPv6 not the solution, it’s part of the problem
 “The really scary thing is that the scaling problem won’t be
obvious until (and if) IPv6 is deployed”—Vince Fuller, 2006
 “Routing and addressing with IPv6 doesn’t really differ from
IPv4—it shares many of the same properties and scaling
characteristics.”—ibid
 Key point: With all current attempts at remediation, routing
complexity scales with the square of the size of the address space
(Dave Meyer, 2009)…
 …Therefore making the address space bigger just makes the
scaling problem worse.
 But that’s a sideshow issue. The real problem is that IP itself is
fatally flawed. The issue isn’t the number after the “v”. It’s with the
letters in front (see next section).


Moore’s Law Also Fails

 The contention: no matter how fast routing tables
grow, Moore’s Law will ensure that the capability of
routers grow faster (transistor density doubles every
2 years)
 This has worked so far.
 So what’s the problem?
 In a nutshell:
 Routing capacity growing roughly at 1.3 X every 2 years
 Routing tables moving towards doubling every two years (2X versus 1.3X)
 Current proposed “solutions” for routing/addressing (including IPv6 and
the loc/ID split) actually accelerate router table growth
 Ergo, Moore’s Law fails!


The Problem: The Devilish Details


History

 For various reasons, the philosophy driving the Internet’s early
development tended to stress pragmatism over architectural
correctness (“rough consensus and running code” was the motto
of the Internet Engineering Task Force)
 The current addressing structure (including IP) was essentially an
extension of the “quick and dirty” addressing structure of the early
Arpanet (which was limited to 64 hosts and did not support
multihoming)
 Even though architecturally-sound approaches were known as
early as 1982, Internet architects continued to expand the address
size without attempting to address known architectural flaws
 Result: IPv4 (4.2 billion addresses and still no multihoming) and
IPv6 (3.4 times 1038 addresses, and still no multihoming)
 Why did this occur? Mix of politics, religion, and expedience


The Devilish Details (Part 1)

 Saltzer showed in 1982 that for an addressing architecture to be
complete, there are three things that need discrete addresses:
 The application
 The node
 The endpoint
 Moreover, there needs to be a defined mapping (directory)
between each layer
 In this architecture, global uniqueness is neither necessary nor
desirable. Each address needs merely to be unambiguous in the
local context.
 That means a routing architecture designed to handle insane
numbers of globally-unique addresses is entirely unnecessary.
The route-scalability problem (which arises in no small part from
the need to route to large numbers of globally-unique addresses)
goes away.
 Oh yeah…And multihoming (and a whole slew of other nice
features) comes for free.


The Devilish Details (Continued)

 IP names an interface, not a node (it’s the electrical socket, not
the toaster)
 The interface isn’t actually a destination. Packets don’t need to go
to an interface, they need to go to a destination.
 But what is a destination? Why, the node! (Actually, the ultimate
destination is an application which is executing on a particular
node at a particular point in time.)
 But in an IP environment, we have failed to define node
addresses. So we talk about electrical sockets when we mean
toasters. (And the statement “IPv6 allows us to address more
devices than there are cells in the human body” is patently false—
IPv6 doesn’t allow us to address the devices at all, just the
interfaces to them.)
 And, oh by the way, we also use IP addresses when what we
mean is application addresses—since we don’t have addresses
for applications, either.


The Gap Between Theory And Practice
 Unfortunately, Saltzer’s architecture is not the one we have with IP.
 In an IP environment, more than half the addressing architecture is
missing. There should be:
 Application addresses
 Node addresses
 Point-of-attachment addresses
 Directories that map between them
 There are:
 No application addresses
 No node addresses
 A point-of-attachment address (MAC address)
 A POA/node interface address (IP address) (See above point about uselessness of
naming interfaces rather than destinations)
 No mappings (DNS maps machine-readable numbers to human-readable addresses. It
doesn’t translate between layers).


The Gap Between Theory And Practice
IP Architecture Correct Architecture

Application
Application Name

Socket (local)

Node Address

IP Address
Point of Attachment
MAC Address Address

Source: John Day, Patterns in Network Architecture


The Devilish Details (Conclusion)

 Q: Why does it matter whether an addressing scheme is
“architecturally complete” if it’s been working pretty well for the past 30
years?
 A: Because it won’t keep working for the next 30 years (or even the
next 5)
 It’s mathematically provable that any architecture that relies on
addressing interfaces instead of nodes will increase route complexity
proportional to the square of the address space
 Therefore, increasing the address space hyper-accelerates the
(already severe) strain on route-table size.
 Eventually routers start dropping routes, and you get:
 Fragmentation: “You can’t get there from here”
 Performance degredation (latency skyrockets)
 Plummeting reliability (roving brownouts)
 Oh, and you still can’t multihome.
 Sound familiar?


Conclusion: Architecture

 An addressing scheme that addresses interfaces
rather than nodes means Internet route complexity will
increase faster than router capability
 Engineering workarounds may keep the problem at
bay for a while, but only by dramatically increasing
costs
 Eventually, routers simply won’t be able to keep up
with demand
 Badness ensues!


The Future

 Machine-to-machine networks
 AT&T now accounts for M2M traffic as a separate revenue category
 User “bandwidth bonding”
 Continued increase in demand
 HD video
 “Always-on” connection pattern
 Increased reliance on wireless
 Embedded intelligent networks…

 …all while facing fundamental technical and regulatory
challenges!


Summary and Takeaways

 The Internet as we know it faces three fundamental
challenges:
 Exponential demand growth
 Access constraints
 Architectural limitations
 To keep it growing and thriving….
 …We’ve got work to do!


Thank You
Questions?
research@nemertes.com

Robin Gareiss
Executive Vice President & Sr. Founding Partner
www.nemertes.com


10 fn key1

Recommended

Recommended

More Related Content

What's hot

What's hot (9)

Similar to 10 fn key1

Similar to 10 fn key1 (20)

More from Scott Foster

More from Scott Foster (20)

10 fn key1