Topologies of Complex Networks Functions vs. Structures Lun Li Advisor: John C. Doyle Co-advisor: Steven H. Low Collaborators: David Alderson (NPS) Walter.

Slides:



Advertisements
Similar presentations
EGEE-III INFSO-RI Enabling Grids for E-sciencE EGEE and gLite are registered trademarks Modeling Grid Job Time Properties Lovro.
Advertisements

Complex Networks Advanced Computer Networks: Part1.
Complex Networks for Representation and Characterization of Images For CS790g Project Bingdong Li 9/23/2009.
CS8803-NS Network Science Fall 2013
Benchmarking traversal operations over graph databases Marek Ciglan 1, Alex Averbuch 2 and Ladialav Hluchý 1 1 Institute of Informatics, Slovak Academy.
HY-483 Presentation On power law relationships of the internet topology A First Principles Approach to Understanding the Internet’s Router-level Topology.
Power Laws By Cameron Megaw 3/11/2013. What is a Power Law?
RESILIENCE NOTIONS FOR SCALE-FREE NETWORKS GUNES ERCAL JOHN MATTA 1.
Synopsis of “Emergence of Scaling in Random Networks”* *Albert-Laszlo Barabasi and Reka Albert, Science, Vol 286, 15 October 1999 Presentation for ENGS.
Internet Topology Caterina Scoglio KSU. Why need for Internet Topology models To evaluate performance of algorithms and protocols Realistic models at.
On Power-Law Relationships of the Internet Topology Michalis Faloutsos Petros Faloutsos Christos Faloutsos.
Weighted networks: analysis, modeling A. Barrat, LPT, Université Paris-Sud, France M. Barthélemy (CEA, France) R. Pastor-Satorras (Barcelona, Spain) A.
Complex Networks Third Lecture TexPoint fonts used in EMF. Read the TexPoint manual before you delete this box.: AA TexPoint fonts used in EMF. Read the.
Trends in Object-Oriented Software Evolution: Investigating Network Properties Alexander Chatzigeorgiou George Melas University of Macedonia Thessaloniki,
Emergence of Scaling in Random Networks Barabasi & Albert Science, 1999 Routing map of the internet
Scale-free networks Péter Kómár Statistical physics seminar 07/10/2008.
The structure of the Internet. How are routers connected? Why should we care? –While communication protocols will work correctly on ANY topology –….they.
Scale Free Networks Robin Coope April Abert-László Barabási, Linked (Perseus, Cambridge, 2002). Réka Albert and AL Barabási,Statistical Mechanics.
Network Design IS250 Spring 2010 John Chuang. 2 Questions  What does the Internet look like? -Why do we care?  Are there any structural invariants?
Common Properties of Real Networks. Erdős-Rényi Random Graphs.
Sampling from Large Graphs. Motivation Our purpose is to analyze and model social networks –An online social network graph is composed of millions of.
From Time Series to Complex Networks B. Luque, L. Lacasa ETSI Aeronáuticos UPM, J. Luque, ETSI Telecomunicaciones UPC, F. Ballesteros, Observatorio Astronómico.
Global topological properties of biological networks.
Graphs and Topology Yao Zhao. Background of Graph A graph is a pair G =(V,E) –Undirected graph and directed graph –Weighted graph and unweighted graph.
Network analysis and applications Sushmita Roy BMI/CS 576 Dec 2 nd, 2014.
Summary from Previous Lecture Real networks: –AS-level N= 12709, M=27384 (Jan 02 data) route-views.oregon-ix.net, hhtp://abroude.ripe.net/ris/rawdata –
Computer Science 1 Web as a graph Anna Karpovsky.
Large-scale organization of metabolic networks Jeong et al. CS 466 Saurabh Sinha.
Information Networks Power Laws and Network Models Lecture 3.
(Social) Networks Analysis III Prof. Dr. Daning Hu Department of Informatics University of Zurich Oct 16th, 2012.
Topic 13 Network Models Credits: C. Faloutsos and J. Leskovec Tutorial
LANGUAGE NETWORKS THE SMALL WORLD OF HUMAN LANGUAGE Akilan Velmurugan Computer Networks – CS 790G.
Network Characterization via Random Walks B. Ribeiro, D. Towsley UMass-Amherst.
Complex Networks First Lecture TexPoint fonts used in EMF. Read the TexPoint manual before you delete this box.: AA TexPoint fonts used in EMF. Read the.
An Optimization-Driven Approach for Modeling AS-level Internet Connectivity Presented by: Hyunseok Chang Joint work with Sugih Jamin.
Emergence of Scaling and Assortative Mixing by Altruism Li Ping The Hong Kong PolyU
Social Network Analysis Prof. Dr. Daning Hu Department of Informatics University of Zurich Mar 5th, 2013.
Challenges and Opportunities Posed by Power Laws in Network Analysis Bruno Ribeiro UMass Amherst MURI REVIEW MEETING Berkeley, 26 th Oct 2011.
Autumn Web Information retrieval (Web IR) Handout #1:Web characteristics Ali Mohammad Zareh Bidoki ECE Department, Yazd University
Yongqin Gao, Greg Madey Computer Science & Engineering Department University of Notre Dame © Copyright 2002~2003 by Serendip Gao, all rights reserved.
Complex Network Theory – An Introduction Niloy Ganguly.
Lecture 10: Network models CS 765: Complex Networks Slides are modified from Networks: Theory and Application by Lada Adamic.
1 Network Intelligence and Networked Intelligence 网络智能和网络化智能 Deyi Li ( 李 德 毅 ) Aug. 1, 2006.
Complex Network Theory – An Introduction Niloy Ganguly.
341- INTRODUCTION TO BIOINFORMATICS Overview of the Course Material 1.
An Improved Acquaintance Immunization Strategy for Complex Network.
School of Information Sciences University of Pittsburgh TELCOM2125: Network Science and Analysis Konstantinos Pelechrinis Spring 2013 Figures are taken.
Network (graph) Models
Lecture 23: Structure of Networks
Random Walk for Similarity Testing in Complex Networks
Linear vs. exponential growth
Empirical analysis of Chinese airport network as a complex weighted network Methodology Section Presented by Di Li.
Marina Leri Institute of Applied Mathematical Research
Normal Distributions.
Lecture 23: Structure of Networks
Self transforming to power law topology for overlay networks.
Biological Networks Analysis Degree Distribution and Network Motifs
Building and Analyzing Genome-Wide Gene Disruption Networks
Peer-to-Peer and Social Networks Fall 2017
Department of Computer Science University of York
Topology and Dynamics of Complex Networks
Peer-to-Peer and Social Networks
From the Synthetic networks slide deck (didn’t have a chance to go over them at that time) By: Ralucca Gera, NPS.
Power Law.
The structure and evolution of scientific collaboration in eLib
Degree Distributions.
Lecture 2: Complex Networks
Lecture 23: Structure of Networks
Lecture 9: Network models CS 765: Complex Networks
Human-centered Machine Learning
Presentation transcript:

Topologies of Complex Networks Functions vs. Structures Lun Li Advisor: John C. Doyle Co-advisor: Steven H. Low Collaborators: David Alderson (NPS) Walter Willinger (AT&T-labs Research)

Incredibly large size Dramatic growth, rapid and ongoing evolution Multitude of measurement data (some are biased and incomplete) Challenges Structures always affect functions Evaluate the performances of new regulations that run on top of the structure. A better design of complex networks Importance

Trends Identify unifying large-scale properties –Power-law degree distribution Build universal models to match properties –Scale-free Networks Derive Emergent properties from models –Achilles heel, self-similar, generic…

Trends Identify unifying large-scale properties –Power-law degree distribution Build universal models to match properties –Scale-free Networks Derive Emergent properties from models –Achilles heel, Self-similar, generic New Science of Networks!

Power Laws Call D={d 1, d 2, …, d n } degree sequence of graph Let d i denote the degree of node i where c>0 and α>0 k : rank of a degree α: power-law tail index, 0<α2 in complex networks Source: Faloutsos et al. (1999) Degree Rank

Power Laws Source: Faloutsos et al. (1999) Most nodes have few connections Call D={d 1, d 2, …, d n } degree sequence of graph Let d i denote the degree of node i A few nodes have lots of connections where c>0 and α>0 Degree Rank

Power-law and High Variability For a sequence D, –CV characterizes the variability of a degree sequence –Regular Graph ( d i =c ), CV(D) = 0 –ER random graph (Poisson), CV(D) = c –Some other random graphs (Exponential), CV(D) = c If D is Power-law (n ), <2, CV(D) = High variability significantly deviates from classical graphs Power-law is discovered in many complex networks Lead to pursue of universal theories to explain it.

Scale-Free (SF) Models Preferential Attachment (PA) Barabasi & Albert (1999) –Growth by sequentially adding new nodes –New nodes connect preferentially to nodes having more connections

Scale-Free (SF) Networks Reproduce power-law degree sequence Generated by random process (PA, GRG, … ) Highly connected central hubs, which are crucial to the system, hold network together –Achilles heel: fragile to specific attack Self-similar and fractal, Small-world properties … SF networks have been suggested as representative models of complex networks

PA GRG

However… Scale-free network theories are incomplete and in need of corrective actions. –Power laws are more normal than Normal… –Power laws are popular but not universal… –And not a signature of specific mechanisms Focus on network functions and structures Functions Internet Router-level topology s-metric Structures

Our Approach for Internet Topology Consider the explicit design of the Internet –Annotated network graphs (bandwidth) –Network Functions Carry expected traffic demand –Constraints Technological constraints Economic limitations –Heuristic optimized tradeoffs (HOT) Maximize network function subject to constraints

Our Approach for Internet Topology Consider the explicit design of the Internet –Annotated network graphs (bandwidth) –Network Functions Carry expected traffic demand –Constraints Technological constraints Economic limitations –Heuristic optimized tradeoffs (HOT)

10K 100K 1M 10M 100M 1G 10G 100G 1000G Total Router Degree (physical connections) Total Router Bandwidth (bits/sec) Shared media at network edge (LAN, DSL, Cable, Wireless, Dial-up) Core backbone High-end gateways Older/cheaper technology Abstracted Technologically Feasible Region Flow conservation in routers: Routers can either have a few high-bandwidth connections, or many low bandwidth connections. Individual router models specialize in different bandwidth- degree combinations and therefore tend to used in different regions of the network.

Rank (number of users) Connection Speed (Mbps) 1e-1 1e-2 1 1e1 1e2 1e3 1e4 1e211e4 1e6 1e8 Dial-up ~56Kbps Broadband Cable/DSL ~500Kbps Ethernet Mbps Ethernet 1-10Gbps most users have low speed connections a few users have very high speed connections high performance computing academic and corporate residential and small business Variability in End-User Bandwidths (2003) High cost of links drives traffic aggregation at network edge

Hosts Edges Core Heuristically Optimal Topology High degree nodes are at the edges. Sparse, mesh-like core of fast, low-degree routers. Relatively uniform low connectivity within core: carry high bandwidth high variability in connectivity at edge: aggregate end users

SOX SFGP/ AMPATH U. Florida U. So. Florida Miss State GigaPoP WiscREN SURFNet Rutgers U. MANLAN Northern Crossroads Mid-Atlantic Crossroads Drexel U.U. Delaware PSC NCNI/MCNCMAGPI UMD NGIX DARPA BossNet GEANT Seattle Sunnyvale Los Angeles Houston Denver Kansas City Indian- apolis Atlanta Wash D.C. Chicago New York OARNET Northern Lights Indiana GigaPoP Merit U. Louisville NYSERNet U. Memphis Great Plains OneNet Arizona St. U. Arizona Qwest LabsUNM Oregon GigaPoP Front Range GigaPoP Texas TechTulane U. North Texas GigaPoP Texas GigaPoP LaNet UT Austin CENICUniNet WIDE AMES NGIX Pacific Northwest GigaPoP U. Hawaii Pacific Wave ESnet TransPAC/APAN Iowa St. Florida A&M UT-SW Med Ctr. NCSA MREN SINet WPI StarLight Intermountain GigaPoP Abilene Backbone Physical Connectivity (as of December 16, 2003) Gbps Gbps Gbps Gbps

Optimization-based models Core: Mesh-like, low degree Edge: High degree From engineering design Tradeoffs in constraints Match the real Internet SF models Core: Hub-like, high degree Edge: Low degree From random process Ignore engineering details Match aggregate statistics SFHOT How to reconcile these two perspectives?

SF PLRG/GRG HOT Abilene-inspiredSub-optimal What are the key differences among these graphs? Functions Structures

Network Performance Given realistic technology constraints on routers, how well is the network able to carry traffic? Step 1: Constrain to be feasible Abstracted Technologically Feasible Region degree Bandwidth (Mbps) Step 3: Compute max flow BiBi BjBj x ij Step 2: Compute traffic demand

SF HOT Perf(g) = 1.19 x (bps) Perf(g) = 1.13 x (bps)

Engineering-based models Core: Mesh-like, low degree Edge: High degree From explicit design Tradeoffs in constraints High throughput High router utilization No Achilles Heel Degree-based models Core: Hub-like, high degree Edge: Low degree From random process Ignore engineering details Low throughput Low router utilization Achilles Heel SF HOT

Engineering-based models Core: Mesh-like, low degree Edge: High degree From explicit design Tradeoffs in constraints High throughput High router utilization No Achilles Heel Degree-based models Core: Hub-like, high degree Edge: Low degree From random process Ignore engineering details Low throughput Low router utilization Achilles Heel PA HOT

A Structural Approach s-metric –Structural metric, depending only on the connectivity of a given graph not on the generation mechanism –Not for a specific network High s(g) is achieved by connecting high degree nodes to each other Measures how hub-like the network core is

s and Joint Degree Distribution Joint Degree Distribution (JDD): p(k,k) correlation between the degrees k, k of connected nodes –Degree distribution is a first order statistic –JDD is a second order statistic For a graph having degree sequence D, s is the aggregation of JDD –Corollary: If two graphs have the same JDD, define a metric as the aggregation of the third order correlation.

P(g) Perfomance (bps) SFHOT S(g)

s-metric Structural metric, depending only on the connectivity of a given graph not on the generation mechanism Define the extent to which the graph is scale- free Differentiate graphs with the same highly variable degree sequence, among them: –s max graph is the one with the highest s value –s min graph is the one with the lowest s value s max – s min defines the graph diversity of a given degree sequence in the simple and connected graph space.

s max s-value s min Variability of a degree sequence Graph diversity cv Variability vs. graph diversity of a degree sequence A tree with 100 nodes, 99 links

s max (SF is closed to s max ) s-value s min (HOT is closed to s min ) Variability vs. graph diversity of a degree sequence A tree with 100 nodes, 99 links Variability of a degree sequence Graph diversity cv

s-value chain star Variability of a degree sequence Graph diversity Low variability graphs cv Variability vs. graph diversity of a degree sequence A tree with 100 nodes, 99 links

s and Assortativity r(g) For a given graph, assortativity is: –r>0, assortative, high degree nodes connect to high degree nodes –r<0, dissortative, high degree nodes connect to low degree nodes –A popular metric to measure the degree correlation

CV(D) Almost all the simple, connected trees with high variable connectivity have negative assortativity Assortativity r max r min Variability vs. assortivity of a degree sequence A tree with 100 nodes, 99 links r(SF) = r(HOT) = -0.46

Assortativity r(g) For a given graph, assortativity is:

Assortativity r(g) For a given graph, assortativity is: Normalization term s max of unconstrained graphs: all the nodes connect themselves

Assortativity r(g) For a given graph, assortativity is: Normalization term Centering term s max of unconstrained graph Center of unconstrained graph

Assortativity r(g) For a given graph, assortativity is: Normalization term Centering term Background set is the unconstrained graph! s max of unconstrained graph Center of unconstrained graph

s-metric assortativity s max s min r max r min CV(D)

s/s max assortativity s min /s max r max r min Assortativity is a metric directed borrowed from classic graph theory It works well for the low variability case Extremely misleading for the high variability complex network CV(D)

s max (SF) and graph metrics With s, we can quantitatively characterize the properties claimed in SF literature. Node Centrality –In s max graph, node centrality increases with degree Small-world phenomena –s max has lowest average shortest path Self similarity –s max graph remains s max by trimming, coarse graining, highest connect motif Generic –s max graph is most likely to appear by GRG

Conclusions Functions Structures The Internet Functions vs. Constraints HOT vs. SF s-metric can highlight the difference (HOT vs SF) s-metric measures graph diversity s-metric has a rich connection to self-similarity, assortativity