220 likes | 231 Views
Explore the economic and structural facets of internet networks, from single to multiple ISP setups. Delve into link-level structures, costs, traffic routing, and growth models, including incremental and global optimization approaches. Investigate how network models evolve with ISP expansion and the implications for service quality and market dynamics.
E N D
Link-Level Internet Structures Scott Schremmer and Kevin Wampler
Overview • Structure of Internet not by chance • Economic factors influence structure • Two areas of focus: • Single ISP structure • Multiple ISP structure
Single ISP Simulation • Focus on network topology • Internet is a graph • Nodes: points of presence • Population served • Geographic location • Edges: physical links • Bandwidth capacity • Need some economic information
Link Cost • Function of distance and capacity • Two components • Cost of cable • Cost of laying cable
Bandwidth Cost • Cost is per unit length cable • Linear in distance • Multiple cables linear in bandwidth • So single cable sublinear in bandwidth • We choose
Traffic Routing • Assume shortest-hop • Distributed evenly amongst population • Traffic between n1 and n2:
Calculating Load • Sum of traffic across link • For each node: • Compute shortest-hop routes • For each target node: • Add traffic to load on all links on route • If load exceeds capacity, excess is overload
Generating Networks • Incremental Growth • Likely a reasonable model • Mimics and ISP adding new POPs • Growth of ISP matters • Global optimization • Entire network planned in advance
Incremental Growth Model • Investigated by Barabasi and Albert • Link probability proportional to degree • Yields power degree distribution • Applicability is unclear • Suboptimal links chosen too often • No direct economic incentive
A Better Model • New nodes added randomly • Optimal link always chosen • Provisioned so no overload • Topology depends on “optimal” • Cost of new link • Cost of new and replaced links • Overprovisioning
Cost Of New Link • Light tailed
Replaced Links • Very heavy tailed
Overprovisioning • Close to power distribution
Global Optimization • Optimize entire network • Growth does not matter • Need objective function • Minimize cost while providing good service • Depends on weighting
Generating A Network • Initially Random • Use Simulated Annealing • Add a link • Remove a link • Grow a link • Shrink a link
Cost/Service Emphasis cost service
Single ISP Conclusions • Topologies similar to reality from: • Minimizing load/link replacement with overprovisioning • Overconnection to improve service
Extension to Multiple ISP’s • Consider that ISPs do not grow in a vacuum • How does model change with introduction of more ISP’s • Link level structures • Economic Choices • Router distributions and routing choices
Specifics of simulation • Simulated annealing • 2 ISP annealed separately • Some number of iterations with one • Pass shared parameters (inter-isp traffic) • Recalculate parameters for annealing separately • Routing-2 models • Small ISP lets large ISP route to cities it doesn’t cover • Small ISP routes to its border
Future work • Consider evolving model • If an isp is providing better service or has lower cost increase its market share • Structures that result • Conditions for isp to survive • Stable? • Anneal together cooperatively, how different are structures • Add nodes to small network to simulate growth