Advanced Computer Networks

Transcription

1 Advanced Computer Networks Network Topology Jianping Pan Summer

2 Motivation Knowing network topology is important network planning traffic engineering performance evaluation protocol design... 2

3 Reality check Network topology is difficult to obtain Internet-scale at application, IP, domain, router levels business disincentive Measuring network topology is difficult AS topology BGP updates: path vector abstraction router topology traceroute alias resolution, measurement bias 3

4 Observations Long-distance links are more expensive Tiered Internet structures locality Waxman random graph (1988) hierarchy GT-ITM structure graph (1996) Power-law degree distribution (1999) preferential attachment degree-based random graph: PA, expected degree 4

5 Power-law distribution At both router and AS levels 5

6 Features of degree-based models Preferential attachment vs expected degree e.g., PLRG, GRG 6

7 New approaches First-principles approach [LAWD04] Lun Li, David Alderson, Walter Willinger, John Doyle. A First-Principles Approach to Understanding the Internet's Router-Level Topology. In SIGCOMM (Best student paper) Follow-on work:d Alderson, L Li, W Willinger, JC Doyle. Understanding Internet Topology: Principles, Models, and Validation. IEEE/ACM TRANSACTIONS ON NETWORKING, Dec

8 First-principles approach Technology constraints Bandwidth (Gbps) Cisco GSR, circa Total Bandwidth 2 1 high BW low degree high degree low BW Bandwidth per Degree 15 x GE 15 x 3 x 1 GE 0 15 x 4 x OC12 15 x 8 FE Technology constraint Degree 2 8 Q: why such constraints?

9 Technology constraints cisco core technologies Total Router BW (Mbps) 0000 approximate aggregate feasible region cisco 124 cisco cisco cisco cisco 7200 older/cheaper technologies 1 linksys 4-port router ubr7246 cmts (cable) cisco 6260 dslam (DSL) edge technologies cisco AS5850 (dialup) degree Source: Cisco Product Catalog, June

10 Economy constraints high performance computing Connection Speed (Mbps) 1e4 Ethernet 1-Gbps 1e3 1e2 academic and corporate a few users have very high speed connections 1e1 1 1e-1 most users have low speed connections 1e-2 1 1e2 residential and small business Ethernet -0Mbps 1e4 Broadband Cable/DSL ~500Kbps Dial-up ~56Kbps 1e6 Rank (number of users) 1e8

11 Evidence: Internet2/Abilene Intermountain GigaPoP Front Range GigaPoP Indiana GigaPoP U. Louisville Great Plains Merit OARNET OneNet Qwest Labs Arizona St. Northern Lights U. Memphis WiscREN NCSA StarLight U. Arizona Iowa St. Oregon GigaPoP Pacific Northwest GigaPoP MREN NYSERNet Pacific Wave UNM Denver Kansas City Chicago Seattle U. Hawaii WIDE GEANT Sunnyvale CENIC SURFNet Wash D.C. Rutgers U. Los Angeles UniNet TransPAC/APAN Houston North Texas GigaPoP Texas Tech Abilene Backbone Physical Connectivity (as of December 16, 2003) Gbps Gbps Gbps Gbps Atlanta SOX Miss State GigaPoP UT Austin UT-SW Med Ctr. MANLAN SFGP/ AMPATH Texas GigaPoP Florida A&M LaNet Tulane U. Northern Crossroads SINet New York ESnet AMES NGIX WPI Indianapolis U. So. Florida MAGPI PSC DARPA BossNet UMD NGIX U. Florida NCNI/MCNC Mid-Atlantic Crossroads Drexel U. U. Delaware 11

12 CENIC CENIC Backbone (as of January 2004) OC-3 (155 Mb/s) OC-12 (622 Mb/s) GE (1 Gb/s) OC-48 (2.5 Gb/s) GE ( Gb/s) Abilene Sunnyvale OAK Cisco 750X Cisco Cisco 124 COR dc2 dc2 SAC hpr hpr FRG dc2 dc3 hpr SVL FRE SOL BAK SLO hpr LAX Abilene Los Angeles hpr dc2 dc3 TUS SDG hpr dc3 12

13 AT&T 13

14 Rocketfuel-inferred AT&T Sprint 14

15 Comparison Degree-based approach follow a given degree distribution power-law: high-degree central hubs First-principles approach technology, economy constraints others: geography, population constraints? Heuristically Optimized Topologies (HOT) faster low-degree backbone networks slower high-degree access networks consistent with real networks 15

16 Same degree distribution... PA PLRG/GRG HOT Abilene-inspired 16 Sub-optimal

17 Network Performance Given realistic technology constraints on routers, how well is the network able to carry traffic? Step 1: Constrain to be feasible Step 2: Compute traffic demand Bj Bandwidth (Mbps) Bi Abstracted Technologically Feasible Region Step 3: Compute max flow max x ij =max αb i B j α degree 1 x ij B i B j xij s. t. i, j i, j : k r ij i, j x ij B k, k 17

18 Implication on network performance Abilene-inspired HOT PA Sub-optimal PLRG/GRG P(g) Perfomance (bps) 12??? 11 Lmax l(g) = 1 P(g) = 1.08 x l(g) = Relative Likelihood

19 This lecture Topology modeling probabilistic random graph structure-based degree-based first-principles technology and economy constraints, reality check impact on network performance Explore further

20 Next lectures July 23: guest lectures July 25, July 30, August 1 Course projects presentation Course evaluation: now! 20

21 One more message... Research opportunities for undergraduates NSERC USRA MITACS internship, on-campus coop directed studies, technical projects Graduate study at UVic UVic CS: systems, theory, applications UVic ECE: communications networks,... financial support NSERC CGS/PGS, BC Pacific Century, UVic Fellowship research and teaching assistantship, coop 21