The Next Generation Internet Program. Mari Maeda ITO

(1)

The Next Generation Internet

Program

The Next Generation Internet

Program

Mari Maeda

ITO

Mari Maeda

ITO

(2)

Today’s Internet

Traffic Makeup

Today’s Internet

Traffic Makeup

HTTP ICMP other

(3)

Today’s Internet

Flow Size Distribution

Comparison of 97 to 99

0 2000 4000 6000 8000 10000 12000 14000 16000 18000 20000 1 74 ₃₃₀

1480 6634 ₂9733

13 3252 59 7196 267 6445 1199 4995

file size (bytes)

Nu m b er o f t ran sf er s Jan. 1999 Jun 1997 Packet Loss -500 500 1500 2500 3500 4500 5500

0 2000 4000 6000 8000

Transmit Rate (kbps)

Pa cke ts Dr oppe d Testbed Internet The Internet.

Cambridge to L.A.

Applications

Digital Video 20-90 Mb High-Definition TV 1500 Mbps

Packet Loss vs. Transmit Rate

Testbed

Application binary 10’s MB High-Resolution Imagery 100 MB to GB

(4)

Scaling the Internet

How do we enable the Internet to scale?

(

in size,speed,reach,apps)

Number of hosts connected to the Internet

1 10 100 1,000 10,000 100,000 1,000,000 10,000,000 100,000,000 1,000,000,000 19 73 19 75 19 77 19 79 19 81 19 83 19 85 19 87 19 89 19 91 19 93 19 95 19 97 19 99 Year Nu mb er

mean hop distance = 16

30 million hosts

• Increased loss probab.delay • delay variation

• decreased security

(5)

Develop next generation multiplexing and

switching technologies that enable dynamic

resource sharing between typical

and high-end users

Supernet

DARPA’s NGI Goals

Create tools that automate planning and mgmt

functions enabling the growth of networks by a

factor of 100 or more, while limiting

the cost and complexity of network

management and control

Network

Engineering

(6)

SuperNet Goals

To enable

ultra-high bandwidth on demand

over national networks, guaranteed over the

shared infrastructure

Target: Multi-Gbps end to end

core network

access network

Approach

Approach::

•

• Streamlined networkingStreamlined networking protocol stacks

protocol stacks

•

• DynamicallyDynamically reconfigurable

reconfigurable/switched/switched

optical layer (opaque or optical layer (opaque or electronic)

electronic)

•

• “Transparency”“Transparency”

•

• New switching/ routingNew switching/ routing technologies and control technologies and control algorithms

algorithms

•

• Dynamic and highDynamic and high

bandwidth local access bandwidth local access

(7)

SuperNet: Simplifying

Protocol Stacks

IP dynamic WDM

•Dynamic bw provisioning •Load balancing WDM WDM WDM ATM ATM ATM SONET SONET SONET SONET SONET IP Router Router Router Router host WDM host ISP1 ISP2

carrier 1 carrier 2

ATM provisioning system provisioning system provisioning system provisioning system 45, 155 Mbps

2.5, 10 Gbps

16,40 λ’s WDM IP ATM SONET Application manually configured

(8)

IP over WDM

•

• WDM based router bypass

WDM based router bypass

•

• Optical Flow Switching -- based on aggregate

Optical Flow Switching -- based on aggregate

traffic change

•

• Host-triggered path setup

Host-triggered path setup

•

• Optical burst switch (v. short holding times)

Optical burst switch (v. short holding times)

WDM WDM

WDM Router

Router Router

Router

host host

WDM

host host

WDM WDM

speed

Dynamic Optical Layer

transparent, opaque, or

regenerated

(9)

IP over WDM

DATA

HEADER

Optical Burst Switch

Optical Label Switching

node1 node2 node3

cntrl

Wavelength

time

data data

(10)

Bitrate and Protocol

Transparent Modules

Bitrate and Protocol

Transparent Modules

‘transparent’ WAN

HD Monitor _Local

Networks

ENG Acquisition

oxc

Modules at the

Modules at the core and the peripherycore and the periphery of the network that can of the network that can •

• Recognize and lock to the bit rate (bit-rate adaptability)Recognize and lock to the bit rate (bit-rate adaptability) •

• Recognize and handle different protocols (protocol agility)Recognize and handle different protocols (protocol agility)

• Dynamically reconfigurable or burst switched networks

• Automated network upgrades without replacing hw (lock-on or sw downloads)

• Rapid deploymet

• Adapt to new types of sensors, CPE’s • Minimum inventory

(11)

À OC3/12/48c ATM / SONET

À OC3/12/48c IP/SONET

À Gigabit ethernet

À SMPTE 25/292

À IEEE 1394 (firewire)

À G-Link

À FDDI

À Fibre Channel

À “ngi protocol” e.g. IP/WDM

Bit-Rate Agile Demux &

Mux

Protocol Processor

Universal Network

Access Module

Universal Network

Access Module

• Target bit range: 100 Mbps to 3 Gbps initially

(10 Gbps later)

• Handle a variety of protocol classes at Layer 1 - 3

• Target bit range: 100 Mbps to 3 Gbps initially

(10 Gbps later)

(12)

Network Engineering

• Adaptive control

• Self-management

• Modeling and simulations

• Network visualization

• Adaptive control

• Self-management

• Modeling and simulations

• Network visualization

(13)

Network Engineering:

Adaptive Network Management Project

Network Engineering:

Adaptive Network Management Project

Self-configuring network monitors

• Surveyors map neighborhood

• They coordinate with other surveyors to adjust their ranges

• Careful multicast based self-organization

– Continuous range expansion – Range description exchange – Back off

• …eventually adapts to surveyor failure, network partitions

Adapts to network fault (link cut,

node failure, congestion, network

partition) and surveyor failure.

Self-configuring network monitors

• Surveyors map neighborhood

• They coordinate with other surveyors

to adjust their ranges

• Careful multicast based

self-organization

– Continuous range expansion

– Range description exchange

– Back off

• …eventually adapts to surveyor

failure, network partitions

Adapts to network fault (link cut,

node failure, congestion, network

partition) and surveyor failure.

Large-scale network fault isolation

Surveyor

(14)

Network Engineering:

Real-Time Network Simulations

Network Engineering:

Real-Time Network Simulations

From:

Off-line

• Yesterday’s traffic situation guides today’s provisioning

• Problems fixed after occurrence

To:

Realtime

• Live parameter tuning • Large-scale changes and

repair validation prior to fielding

simulators

real world networks

parameter tuning topology

/configuration

(15)

Adaptive Web Caching Project

Target Problem: “Hot Spots”

Adaptive Web Caching Project

Target Problem: “Hot Spots”

Hundreds of thousands of clients fetching the same data

from the same server at about the same time

Hundreds of thousands of clients fetching the same data

from the same server at about the same time

Today:

• Happens few times a year • Manually create replic. sites • The Internet has yet to meet the

challenge of simultaneous demands from millions of users

Tomorrow:

• Daily occurrence?

• Need demand-driven data

dissemination and self-organizing caches e.g. content based routing protocol, cache group management protocol

(16)

Network Engineering: Network

Monitoring, Analysis and Visualization

Network Engineering: Network

Monitoring, Analysis and Visualization

• Monitor and automate the discovery of the

topology and traffic behavior of the

Internet and future networks on a global

scale.

• What makes this hard:

À No central authority

À Scale (span and speed)

À Capturing dynamic behavior

À Visualization

Tools

:

“skitter” (active measurements: performance, topology)

“coral” monitors (passive measurements over high speed links)

• Monitor and automate the discovery of the

topology and traffic behavior of the

Internet and future networks on a global

scale.

• What makes this hard:

À No central authority

À Scale (span and speed)

À Capturing dynamic behavior

À Visualization

Tools

:

“skitter” (active measurements: performance, topology)

(17)

UCSD/CAIDA

(Cooperative Association for Internet Data Analysis)

Network Tomography

• Network “Radar”: Global

connectivity information

• Measure IP paths (“hops”)

from source to MANY (~104)

destinations

• Use 52 byte ICMP echo

requests (every 30 min.) as

probes

• Challenges:

– Pervasive measurement with minimal load on infrastructure – Visualization

• Network “Radar”: Global

connectivity information

• Measure IP paths (“hops”)

from source to MANY (~104)

destinations

• Use 52 byte ICMP echo

requests (every 30 min.) as

probes

• Challenges:

– Pervasive measurement with minimal load on infrastructure – Visualization

(18)

Internet Tomography

Hop count

histogram

Temporal

behavior

(19)

HSCC 2.5 Gb/s BossNet dark fibers Boston D.C. NTON II NTON II 4 wavelengths 4 wavelengths

@ 10 Gb/s per

@ 10 Gb/s perλλ

Seattle NASA/ Ames NASA/ Ames SNL SNL UC Berkeley LBNL LBNL vBNS vBNS SRI SRI BART Sprint LLNL ACTS ACTS ACTS Los Angeles San Diego Portland San Francisco NSA NRL NIMA DARPA DISA _DIA NASA