the role of design in the internet and other complex systems
DESCRIPTION
The Role of Design in the Internet and Other Complex Systems. David Alderson February 10, 2004 Joint work with J. Doyle, W. Willinger, and L. Li. My challenge. Use models of Internet topology as a case study to illustrate many of the themes of this week - PowerPoint PPT PresentationTRANSCRIPT
The Role of Design in the Internet and Other Complex Systems
David AldersonFebruary 10, 2004
Joint work with J. Doyle, W. Willinger, and L. Li
My challengeUse models of Internet topology as a case study to
illustrate many of the themes of this week– “How to make complex systems still complex but
experimentally accessible?”– Importance/interpretation of high variability in
complex systems– Modeling debate: design vs. randomness– Understanding the “robust, yet fragile” aspects of the
Internet– “Closing the loop” between modeling and analysis– Similarity to models in biology?
The Internet as a Case Study
• To the user, it creates the illusion of a simple, robust, homogeneous resource enabling endless varieties and types of technologies, physical infrastructures, virtual networks, and applications (heterogeneous).
• Its complexity is starting to approach that of simple biological systems
• Our understanding of the underlying technology together with the ability to perform detailed measurements means that most conjectures about its large-scale properties can be unambiguously resolved, though often not without substantial effort.
IP
TCP/AQM
Applications
Link
Source coding
FAST TCP/AQM
IP routing
HOT topology
A Theory for the Internet?
General Approach:Use an engineering design perspective
to understand, explainthe complex structure observed.
Take a single layer in isolation and assume thatthe other layers are handled near optimally.
IP
TCP/AQM
Applications
Link
A Theory for the Internet?
If TCP/AQM is the answer, what is the
question?
Primal/dual model of TCP/AQM congestion control…
cRx
xUs
ssxs
subject to
)( max0
gives
gives
??
IP
TCP/AQM
Applications
Link
A Theory for the Internet?
??If the current topology of the Internet is the answer,
what is the question?
The Internet hourglass
IP
Web FTP Mail News Video Audio ping napster
Applications
TCP SCTP UDP ICMP
Transport protocols
Ethernet 802.11 SatelliteOpticalPower lines BluetoothATM
Link technologies
IP
The Internet hourglass
Web FTP Mail News Video Audio ping napster
Applications
TCP
Ethernet 802.11 SatelliteOpticalPower lines BluetoothATM
Link technologies
The Internet hourglass
IP
Web FTP Mail News Video Audio ping napster
Applications
TCP
Ethernet 802.11 SatelliteOpticalPower lines BluetoothATM
Link technologies
Everythingon IP
IP oneverything
Network protocols.
HTTP
TCP
IP
Files
packetspacketspacketspacketspacketspackets
Sources
Links
Files
Sources
Links
Hosts
Routers
Sources
Links
Sources
Links
Hosts
Routers
packets
Modeling Network TopologyWhy does it matter?
1. Performance evaluation of protocols
2. Provisioning• Topology constrains the applications and services
that run on top of it
3. Understanding large-scale properties • Reliability and robustness to accidents, failures,
and attacks on network components
4. Insight into other network systems• To the extent that the network model is “universal”
Topology Modeling
• Direct inspection generally not possible
• Recent trend: generative models follow empirical measurement studies
• But…– So many things to measure– Incredible variability in so many aspects– How to determine what matters?
The Internet• Full of “high variability”
– Link bandwidth: Kbps – Gbps– File sizes: a few bytes – Mega/Gigabytes– Flows: a few packets – 100,000+ packets– In/out-degree (Web graph): 1 – 100,000+– Delay: Milliseconds – seconds and beyond
• How should we think about the incredible scaling ability of the Internet?
• Is there something “universal” about its structure?
Topology Modeling
• Direct inspection generally not possible
• Recent trend: generative models follow empirical measurement studies
• But…– So many things to measure– Incredible variability in so many aspects– How to determine what matters?
• We will focus on router-level topology
Router-Level Topology
Hosts
Routers
• Nodes are machines (routers or hosts) running IP protocol
• Measurements taken from traceroute experiments that infer topology from traffic sent over network
• Subject to sampling errors and bias
• Requires careful interpretation
Power Laws and Internet Topology
Source: Faloutsos et al (1999)
• How to account for high variability in node degree?• Can we develop an explanatory model for the current
network topology?
Most nodes have few connections
A few nodes have lots of connections
Power laws are ubiquitous• This is no surprise, and requires no “special”
explanation. • Gaussians (“Normal”) distributions are attractors
for averaging (e.g Central Limit Theorem) so are also ubiquitous.
• Power laws are attractors for averaging too, but are also the only distributions invariant under maximizing, marginalization, and mixtures.
• For high variability data subject to these operations, power laws should be expected (Power laws as “more normal than Normal”?)
20th Century’s 100 largest disasters worldwide
10-2
10-1
100
100
101
102
US Power outages (10M of customers)
Natural ($100B)
Technological ($10B)
Log(size)
Log(rank)
100
101
102
1
2
3
10
100
10-2
10-1
100
Log(size)
Log(rank)
100
101
102
20th Century’s 100 largest disasters worldwide
US Power outages (10M of customers,1985-1997)
Natural ($100B)
Technological ($10B)
Slope = -1(=1)
10-2
10-1
100
? 10
0
101
102 US Power outages
(10M of customers, 1985-1997)
10-2
10-1
100
Slope = -1(=1)
A large event is not inconsistent with statistics.
Our Perspective• Must consider the explicit design of the Internet
– Protocol layers on top of a physical infrastructure– Physical infrastructure constrained by technological
and economic limitations– Emphasis on network performance– Critical role of feedback at all levels
• We seek a theory for Internet topology that is explanatory and not merely descriptive.
• Consider the ability to match large scale statistics (e.g. power laws) as secondary evidence of having accounted for key factors affecting design
HOTHighly HeavilyHeuristically
Optimized Organized
Tolerance Tradeoffs
• Based on ideas of Carlson and Doyle• Complex structure (including power laws) of highly
engineered technology (and biological) systems is viewed as the natural by-product of tradeoffs between system-specific objectives and constraints
• Non-generic, highly engineered configurations are extremely unlikely to occur by chance
Heuristic Network DesignWhat factors dominate network design?
• Economic constraints – User demands – Link costs– Equipment costs
• Technology constraints – Router capacity– Link capacity
1e-1
1e-2
1
1e1
1e2
1e3
1e4
1e21 1e4 1e6 1e8
Rank (number of users)
Dial-up~56Kbps
BroadbandCable/DSL~500Kbps
Ethernet10-100Mbps
POS/Ethernet1-10Gbps
Con
nec
tion
Sp
eed
(M
bp
s)
most users have low speed
connections
a few users have very high speed
connections
high performancecomputing
academic and corporate
residential and small business
Internet End-User Bandwidths
How to build a network that
satisfies these end user demands?
Economic Constraints• Network operators have a limited budget to
construct and maintain their networks
• Links are tremendously expensive
• Tremendous drive to operate network so that traffic shares the same links– Enabling technology: multiplexing– Resulting feature: traffic aggregation at edges– Diversity of technologies at network edge (Ethernet,
DSL, broadband cable, wireless) is evidence of the drive to provide connectivity and aggregation using many media types
Heuristically Optimal Network
Hosts
Edges
CoresMesh-like core of fast,
low degree routers
High degree nodes are at the edges.
Heuristically Optimal NetworkClaim: economic considerations alone yield
• Mesh-like core of high-speed, low degree routers
• High degree, low-speed nodes at the edge
• Is this consistent with technology capability?
• Is this consistent with real network design?
Cisco 12000 Series Routers
Chassis Rack size SlotsSwitching Capacity
12416 Full 16 320 Gbps
12410 1/2 10 200 Gbps
12406 1/4 6 120 Gbps
12404 1/8 4 80 Gbps
• Modular in design, creating flexibility in configuration.
• Router capacity is constrained by the number and speed of line cards inserted in each slot.
Source: www.cisco.com
Cisco 12000 Series RoutersTechnology constrains the number and capacity of line cards that can be installed, creating a feasible region.
Cisco 12000 Series RoutersPricing info: State of Washington Master Contract, June 2002
(http://techmall.dis.wa.gov/master_contracts/intranet/routers_switches.asp)
$602,500
$381,500
$212,400
$128,500
$2,762,500
$1,667,500
$932,400
$560,500
bandwidth
degree1 16
10Gb
155Mb
256log/log
625Mb
2.5GbTechnically
feasible
160Gb
Technological advance
Technologically Feasible Region
0.01
0.1
1
10
100
1000
10000
100000
1000000
1 10 100 1000 10000
degree
Ban
dw
idth
(M
bp
s) cisco 12416
cisco 12410
cisco 12406
cisco 12404
cisco 7500
cisco 7200
cisco 3600/3700
cisco 2600
linksys 4-port router
uBR7246 cmts(cable)cisco 6260 dslam(DSL)cisco AS5850(dialup)
Edge Shared media(LAN, DSL,
Cable, Wireless,Dial-up)
Corebackbone
High-end gateways
Older/cheapertechnology
Sprint backbone
SOX
SFGP/AMPATH
U. Florida
U. So. Florida
Miss StateGigaPoP
WiscREN
SURFNet
Rutgers
MANLAN
NorthernCrossroads
Mid-AtlanticCrossroads
Drexel
U. Delaware
PSC
NCNI/MCNC
MAGPI
UMD NGIX
DARPABossNet
GEANT
Seattle
Sunnyvale
Los Angeles
Houston
Denver
KansasCity
Indian-apolis
Atlanta
Wash D.C.
Chicago
New York
OARNET
Northern LightsIndiana GigaPoP
MeritU. Louisville
NYSERNet
U. Memphis
Great Plains
OneNetArizona St.
U. Arizona
Qwest Labs
UNM
OregonGigaPoP
Front RangeGigaPoP
Texas Tech
Tulane U.
North TexasGigaPoP
TexasGigaPoP
LaNet
UT Austin
CENICUniNet
WIDE
AMES NGIX
OC-3 (155 Mb/s)OC-12 (622 Mb/s)GE (1 Gb/s)OC-48 (2.5 Gb/s)OC-192/10GE (10 Gb/s)
Abilene BackbonePhysical Connectivity(as of December 16, 2003)
PacificNorthwestGigaPoP
U. Hawaii
PacificWave
ESnet
TransPAC/APAN
Iowa St.
Florida A&MUT-SWMed Ctr.
NCSA
MREN
SINet
WPI
StarLight
IntermountainGigaPoP
Cisco 750X
Cisco 12008
Cisco 12410
dc1dc2
dc3
hpr
dc1
dc3
hpr
dc2dc1
dc1dc2
hprhpr
SACOAK
SVL
LAX
SDG
SLOdc1
FRGdc1
FREdc1
BAKdc1
TUSdc1
SOLdc1
CORdc1
hprdc1
dc2
dc3
hpr
OC-3 (155 Mb/s)OC-12 (622 Mb/s)GE (1 Gb/s)OC-48 (2.5 Gb/s)10GE (10 Gb/s)
CENIC Backbone (as of January 2004)
AbileneSunnyvale
AbileneLos Angeles
Backbone topology of both Abilene and CENIC are both built as a mesh of high speed, low degree routers.
As one moves from the core out toward the edge, connectivity gets higher, and speeds get lower.
dc1dc2
dc3
hpr
dc1
dc3
hpr
LAX
SDG
SLOdc1
BAKdc1
TUSdc1
hpr
Chaffey, Crafton Hills, Cypress, Fullerton CC,
Mt. San Jacinto, Rio Hondo, Riverside, San Bernardino CCD, San Bernardino Valley, N.
Orange Cty CCD, Santa Ana College
Chaffey Joint USD
LA USD
UC Irvine
UC San Diego
SDSC
UCLA
UC Riverside
San Diego CC, Soutwestern CC,
Grossmont, Cuyamaca, Imperial Valley, Mira Costa CC, Palomar
CollegeSan Diego COE Johnson & Johnson
CUDI Peer,ESNet Peer
LosNettos
LAAP
UCSSN(Las Vegas)
UC SantaBarbara
MonroviaUSD Gigaman
Antelope Valley CC, Cerritos, Citrus, College of
the Canyons, Compton, East LA, El Camino CC, Glendale, Long Beach City College, Pasadena
CC, Santa Monica, Ventura
College
LA CCD, LA City, LA Harbor, LA Mission, LA Pierce, LA Southwest, LA Trade Tech,
LA Valley, Moorpark, Mt. San Antonio, Oxnard
Los Angeles COE
San Bernardino CSS
Riverside COE
Orange COE
Caltech
Abilene
to Soledad
to Sunnyvale
to Sacramento
to Fremont
CENIC Backbone for Southern California
Heuristically Optimal Network• Mesh-like core of high-speed, low degree routers• High degree, low-speed nodes at the edge
• Claim: consistent with drivers of topology design– Economic considerations (traffic aggregation)
– End user demands
• Claim: consistent with technology constraints• Claim: consistent with real observed networks
Question: How could anyone imagine anything else?
Two opposite views of complexity
Physics:• Pattern formation by
reaction/diffusion• Edge-of-chaos• Order for free• Self-organized criticality• Phase transitions• Scale-free networks• Equilibrium, linear• Nonlinear, heavy tails as
exotica
Engineering and math:• Constraints• Tradeoffs• Structure • Organization• Optimality• Robustness/fragility• Verification• Far from equilibrium• Nonlinear, heavy tails as
tool
Models of Internet Topology• Random graphs [Waxman ’88]• Explicit hierarchy [Calvert/Zegura ’96]• Power laws [Faloutsos3 ’99]
Random NetworksTwo methods for generating random networks having
power law distributions in node degree• Preferential attachment (“scale-free” networks)
– Inspired by statistical physics– Barabasi et al.; 1999
• Power Law Random Graph (PLRG)– Inspired by graph theory– Aiello, Chung, and Lu; 2000
Common features:• Ignore all system-specific details• Central core of high-degree, hub-like nodes
Summary of Scale-Free Story• Fact: Scale-free networks have roughly power law
degree distributions• Claim:
– If the Internet has power law degree distribution– Then it must be scale-free (oops)– Therefore, it has the properties of a scale-free network
One ofthe most-read papers ever on
the Internet!
Scientists spot Achilles heel of the Internet
• "The reason this is so is because there are a couple of very big nodes and all messages are going through them. But if someone maliciously takes down the biggest nodes you can harm the system in incredible ways. You can very easily destroy the function of the Internet," he added.
• Barabasi, whose research is published in the science journal Nature, compared the structure of the Internet to the airline network of the United States.
Complexity Digest 2004.06 Feb. 09, 2004Archive: http://www.comdig.org
13. Accurately Modeling the Internet Topology , arXiv
Abstract: To model the behavior of a network it is crucial to obtain a good destopology because structure affects function. When studying the topological propInternet, we found out that there are two mechanisms which are necessary for thof the Internet: a nonlinear preferential growth, where the growth is describedpositive-feedback mechanism, and the appearance of new links between already exshow that the Positive-Feedback Preference (PFP) model, which is based on the areproduces topological properties of the Internet such as: degree distribution,(rich-club connectivity), shortest path length, neighbor clustering, network reand rectangle coefficient), disassortative mixing (nearest-neighbors average deinformation flow pattern (betweenness centrality). We believe that these growthfurther study because they provide a novel insight into the evolutionary dynamics ofnetworks.
* [38] Accurately Modeling the Internet Topology, Shi Zhou, Raul J. Mondragon, , 2004-02-05, arXiv
Key Points• The scale-free story is based critically on the implied
relationship between power laws and a network structure that has highly connected “central hubs”– Not all networks with power law degree distributions
have properties of scale free networks. (The Internet is just one example!)
– Building a model to replicate power law data is no more than curve fitting (descriptive, not explanatory)
• The scale-free models ignore all system-specific details in making their claims– Ignore architecture (e.g. hardware, protocol stack)
– Ignore objectives (e.g. performance)
– Ignore constraints (e.g. geography, economics)
End Result
The scale-free claims of the Internet are not merely wrong, they suggest properties that are the opposite
of the real thing.
Fundamental difference:
random vs. designed
Internet topologies
nodes=routersedges=links
25 interior routers818 end systems
High degree hub-like coreLow degree
mesh-like core
101
102
100
101
degree
rank
identical power-law degrees
How to characterize / compare these two networks?
“scale-rich” vs. scale-free
Network PerformanceGiven realistic technology constraints on routers, how
well is the network able to carry traffic?
Step 1: Constrain to be feasible
Abstracted Technologically Feasible Region
1
10
100
1000
10000
100000
1000000
10 100 1000
degree
Ban
dw
idth
(M
bp
s)
Bi
Bj
xij
Step 2: Compute traffic demand
kBxts
BBx
ijrkjikij
ji jijiij
,..
maxmax
:,
, ,
Step 3: Compute max flow
Network LikelihoodHow likely is a particular graph (having given node
degree distribution) to be constructed?
• Notion of likelihood depends on defining an appropriate probability space for random graphs.
• Many methods (all based on probabilistic preferential attachment) for randomly generating graphs having power law degree distributions:– Power Law Random Graph (PLRG) [Aiello et al.]
– Random rewiring (Markov chains)
In both cases, LogLikelihood (LLH) j
connectedji
idd,
Likelihood
HighLow
Fast
Slow
Performance
Why such strikingdifferences with same
node degree distribution?
1
10
100
1000
10000
100000
10 100
degree
Ban
dw
idth
(M
bp
s)
11
10
100
1000
10000
100000
10 100
degree1
Fast core
High-degree edge
Slow core
Slower edge
Performance LikelihoodLikelihood
HighLow
Fast
Slow
Scale-free• Core: Hub-like, high degree • Edge: Low degree• Robust to random • Fragile to “attack”
HOT scale-rich• Core: Mesh-like, low degree • Edge: High degree• Robust to random • Robust to “attack”
• High performance• Low link costs• Unlikely, rare, designed• Destroyed by rewiring• Similar to real Internet
• Low performance• High link costs• Highly likely, generic• Preserved by rewiring• Opposite of real Internet
+ objectives and constraints
HierarchicalScale-Free (HSF)
RandomHOT
Low LikelihoodLow Performance Most Likely
LikelihoodHighLow
Robust, Efficient
Fragile, Wasteful
scale-free, critical, SOC, edge-of-chaos
The only functional biological or technological networks are highly organized, robust, efficient, and very unlikely to arise by random.
HOT
Universal features of complex networks
HOT=Highly Organized/Optimized Tradeoffs/Tolerance
1 10 100
100
101
102
103
Number of reactions
Rank
Carriers
all metabolites
Reactions
Metabolites
58 133 190 240
33
65
78
132
152
184
204
236
251
313
Carriers
Catabolism
Amino acids
Nucleotides
Lipids & fatty acids
Cofactors
Precursors
H. Pylori
CarriersAmino acids
Nutrients
Precursors
Nucleotides
Fatty acidsAnd Lipids
Cofactors
Bowtie architecture
Catabolism
Biosynthesis
1 2
3 4
S ATP S ADP
S NADH S NAD
Stoichiometry
S1 S2
S3 S4
NADNADH
ADPATP
Substrate
Carrier
Reaction,Enzyme
StoichiometryMatrix
1
2
3
4
1 0
1 0
0 1
0 1
1 0
1 0
0 1
0 1
S
S
S
S
ATP
ADP
NADH
NAD
GLC
DPG
PGL
PGC RL5P
X5P
6PG
MAL
CIT
ICIT
SUCFUM
G6P
F6P
T3P
3PG
PEP
R5P
E4P
PYR
OA
AKG
SUCOA
PRPP
DAH DQT DHS SME S5P PSM CHO
AN NAN CD5 IGP
PPN HPP
BAPASS
HSE PHSDHD
PIP SAK SDP DPI MDP
PHP PPS
ASE
GLN
GLU
SER
TRP
ASP
TYR
THR
LYS
CYS
GLY
ASN
GLC
DPG
PGL
PGC RL5P
X5P
6PG
MAL
CIT
ICIT
SUCFUM
G6P
F6P
T3P
3PG
PEP
R5P
E4P
PYR
OA
AKG
SUCOA
PRPP
DAH DQT DHS SME S5P PSM CHO
AN NAN CD5 IGP
PPN HPP
BAPASS
HSE PHSDHD
PIP SAK SDP DPI MDP
PHP PPS
ASE
GLN
GLU
SER
TRP
ASP
TYR
THR
LYS
CYS
GLY
ASN
PI NADNADH
ADPATP
NADPNADPH CO2 COA
ACCOA PPI AMPATP NH3 AC THF
MTH H2S
PI NADNADH ADPATP NADPNADPH CO2 COAACCOA PPI AMPATP NH3 AC THFMTH H2S
Carriers
precursors amino acids
• WT is highly organized, structured
• Simple reactions• Long assembly lines• Universal common carriers• Precursors and carriers are universal common currenciesWild type
GLC
DPG
PGL
PGC RL5P
X5P
6PG
MAL
CIT
ICIT
SUCFUM
G6P
F6P
T3P
3PG
PEP
R5P
E4P
PYR
OA
AKG
SUCOA
PRPP
DAH DQT DHS SME S5P PSM CHO
AN NAN CD5 IGP
PPN HPP
BAPASS
HSE PHSDHD
PIP SAK SDP DPI MDP
PHP PPS
ASE
GLN
GLU
SER
TRP
ASP
TYR
THR
LYS
CYS
GLY
ASN
PI NADNADH ADPATP NADPNADPH CO2 COAACCOA PPI AMPATP NH3 AC THFMTH H2S
Carriers
precursors amino acids
GLC
DPG
PGL
PGC RL5P
X5P
6PG
MAL
CIT
ICIT
SUCFUM
G6P
F6P
T3P
3PG
PEP
R5P
E4P
PYR
OA
AKG
SUCOA
PRPP
DAH DQT DHS SME S5P PSM CHO
AN NAN CD5 IGP
PPN HPP
BAP
ASSHSE PHS
DHDPIP SAK SDP DPI MDP
PHP PPS
ASE
GLN
GLU
SER
TRP
ASP
TYR
THR
LYS
CYS
GLY
ASN
PI NADNADH ADPATP NADPNADPH CO2 COAACCOA PPI AMPATP NH3 AC THFMTH H2S• Randomly rewire to get “scale-free” version• Preserve
• degree• carrier and enzyme
• Destroys structure• Only one useful pathway remains
Random
GLC
DPG
PGL
PGC RL5P
X5P
6PG
MAL
CIT
ICIT
SUCFUM
G6P
F6P
T3P
3PG
PEP
R5P
E4P
PYR
OA
AKG
SUCOA
PRPP
DAH DQT DHS SME S5P PSM CHO
AN NAN CD5 IGP
PPN HPP
BAPASS
HSE PHSDHD
PIP SAK SDP DPI MDP
PHP PPS
ASE
GLN
GLU
SER
TRP
ASP
TYR
THR
LYS
CYS
GLY
ASN
PI NADNADH ADPATP NADPNADPH CO2 COAACCOA PPI AMPATP NH3 AC THFMTH H2S
Carriers
precursors amino acids
GLC
DPG
PGL
PGC RL5P
X5P
6PG
MAL
CIT
ICIT
SUCFUM
G6P
F6P
T3P
3PG
PEP
R5P
E4P
PYR
OA
AKG
SUCOA
PRPP
DAH DQT DHS SME S5P PSM CHO
AN NAN CD5 IGP
PPN HPP
BAP
ASSHSE PHS
DHDPIP SAK SDP DPI MDP
PHP PPS
ASE
GLN
GLU
SER
TRP
ASP
TYR
THR
LYS
CYS
GLY
ASN
PI NADNADH ADPATP NADPNADPH CO2 COAACCOA PPI AMPATP NH3 AC THFMTH H2S
GLC
DPG
PGL
PGC RL5P
X5P
6PG
MAL
CIT
ICIT
SUCFUM
G6P
F6P
T3P
3PG
PEP
R5P
E4P
PYR
OA
AKG
SUCOA
PRPP
DAH DQT DHS SME S5P PSM CHO
AN NAN CD5 IGP
PPN HPP
BAPASS
HSE PHSDHD
PIP SAK SDP DPI MDP
PHP PPS
ASE
GLN
GLU
SER
TRP
ASP
TYR
THR
LYS
CYS
GLY
ASN
PI NADNADH
ADPATP
NADPNADPH CO2 COA
ACCOA PPI AMPATP NH3 AC THF
MTH H2S
Random Wild type
GLC
DPG
PGL
PGC RL5P
X5P
6PG
MAL
CIT
ICIT
SUCFUM
G6P
F6P
T3P
3PG
PEP
R5P
E4P
PYR
OA
AKG
SUCOA
PRPP
DAH DQT DHS SME S5P PSM CHO
AN NAN CD5 IGP
PPN HPP
BAP
ASS
HSE PHSDHD
PIP SAK SDP DPI MDP
PHP PPS
ASE
GLN
GLU
SER
TRP
ASP
TYR
THR
LYS
CYS
GLY
ASN
PI NADNADH
ADPATP
NADPNADPH CO2 COA
ACCOA PPI AMPATP NH3 AC THF
MTH H2S
Modeling Analysis
ValidationMeasurement
“Closing the loop”
PreferentialAttachment
PLRG
HOT
PreferentialAttachment PLRG HOT
PreferentialAttachment PLRG HOT
1e6
1e5
1
1e1
1e2
1e3
1e4
1e11 1e2 1e3 1e4
Degree (number of connections)
Tot
al R
oute
r B
and
wid
th (
Mb
ps)
Core Routers
High-EndGateways
AccessEdge RoutersShared Media
OlderCheaper
Technology
AbstractedFeasible Region
Internet Routing Technologies
Degree (number of connections)
1e6
1e5
1
1e1
1e2
1e3
1e4
1e11 1e2 1e3 1e4
Tot
al R
oute
r B
and
wid
th (
Mb
ps)
Core Routers High-End
Gateways
AccessEdge RoutersShared Media
OlderCheaper
Technology
AbstractedFeasible Region
Internet Routing Technologies
Per link bandwidth
1e-1
1e-2
1
1e1
1e2
1e3
1e4
1e11 1e2 1e3 1e4
Degree (number of connections)
Ban
dw
idth
/ L
ink
(M
bp
s)
Dial-up~56Kbps
BroadbandCable
~500Kbps
DSL~500Kbps
Core Routers10Gbps Core/Edge
Routers1Gbps
Local AreaEthernet
10-100Mbps
Internet Link Speeds