Network Topology: ELEG 667-013 Spring 2003
Network Topology: ELEG 667-013 Spring 2003
Network Topology: ELEG 667-013 Spring 2003
Outline:
Why Network Topology is Important ? Modeling Internet Topology Complex Networks Scale-free Networks Power-laws of the Web Search in power-law networks: GNUTELLA, a P2P example.
exchange point
border routers
ki (ki ) jk
Many extensions.
Waxman Model:
Router level model Nodes placed at random in 2D space with dimension L Probability of edge (u,v):
d(u,v)
Transit-Stub Model:
Router level model Transit domains placed in 2D space populated with routers connected to each other Stub domains placed in 2D space populated with routers connected to transit domains Models hierarchy Edge count, guaranteed connectivity
Transit-Stub Model:
No concept of a host all nodes are routers. Two level hierarchy First generate a number of transit domains, then generate a set of stub networks. Given average edge-count, produce a random graph, making sure that it is connected.
Inet:
Generate degree sequence Build spanning tree over nodes with degree larger than 1, using preferential connectivity randomly select node u not in tree join u to existing node v with probability d(v)/d(w) Connect degree 1 nodes using preferential connectivity Add remaining edges using preferential connectivity
BRITE:
Generate small backbone, with nodes placed: randomly or concentrated (skewed) Add nodes one at a time (incremental growth) New node has constant # of edges connected using: preferential connectivity and/or locality
Complex Networks:
Two limiting-case topologies have been extensively considered in the literature [4],[5].: regular network (lattice), the chosen topology of innumerable physical models such as the Ising model or percolation. random graph, studied in mathematics and used both in natural and social sciences. Properties studied in detail by Pal Erdos. Most of Erdos work concentrated on the case in which the number of vertices is kept constant but the total number of links between vertices increases: the Erds-Rnyi result states that for many important quantities there is a percolation-like transition at a specific value of the average number of links per vertex.
Complex Networks:
random networks are used in: Physics: in studies of dynamical problems, spin models and thermodynamics, random walks, and quantum chaos. Economics and social sciences: to model interacting agents.
Complex Networks:
In contrast to these two limiting topologies, empirical evidence suggests that many biological, technological or social networks appear to be somewhere in between these extremes. many real networks seem to share with regular networks the concept of neighborhood, which means that if vertices i and j are neighbors then they will have many common neighbors --- which is obviously not true for a random network. On the other hand, studies on epidemics show that it can take only a few ``steps'' on the network to reach a given vertex from any other vertex. This is the foremost property of random networks, which is not fulfilled by regular networks.
Complex Networks:
Complex Networks:
The Watts-Strogatz model [5]. : To bridge the two limiting cases, Watts and Strogatz [Nature 393, 440 (1998)] have introduced a new type of network which is obtained by randomizing a fraction p of the links of the regular network. Initial structure (p=0) is the one-dimensional regular network where each vertex is connected to its z nearest neighbors. For 0 < p < 1, we denote these networks disordered. for the case p=1, we have a completely random network.
Complex Networks:
Watts and Strogatz report that for a small value of the parameter p, there is an onset of small-world behavior. It is characterized by the fact that the distance between any two vertices is of the order of that for a random network and, at the same time, the concept of neighborhood is preserved. The effect of a change in p is extremely nonlinear, where a very small change in the connectivity of the network leads to a dramatic change in the distance between different pairs of vertices.
Complex Networks:
The scientific question we are trying to answer is: Does the onset of the small-world behavior occurs at a given value of p or does it occur for a value of the system size n which depends on p? To investigate this question, we need to look at the behavior of the system as a function of p for different values of n.
Complex Networks:
Complex Networks:
The appearance of the small-world behavior is not a phasetransition but a crossover phenomena. The average distance l is: l (n,p) ~ n* F ( n / n* ) where: F(u << 1) ~ u, and F(u >> 1) ~ln u, and n* is a function of p.
When the average number of rewired links, pnz/2, is much less than one, the network should be in the large-world regime. On the other hand, when pnz/2 >> 1, the network should be a small-world.
Scale-free networks:
It was proposed by Barabsi and Albert that real-world networks in general are scale-free networks. Scale-free networks have a distribution of connectivities that decays with a power-law tail. Scale-free networks emerge in the context of a growing network in which new vertices connect preferentially to the more highly connected vertices in the network. Scale free networks are also small-world networks because (i) they have clustering coefficients much larger than random networks, and (ii) their diameter increases logarithmically with the number of vertices n.
P(k ) k
Characteristic property of Scale free networks Occur very often in Complex Systems literature. Many complicated real world networks obey power laws
Scale-free networks:
Scale-free networks: (a) the neuronal network of the worm C. elegans. (b) world-wide web. (c) the network of citations of scientific papers.
Scale-free networks:
broad-scale networks: or truncated scale-free networks, characterized by a connectivity distribution that has a powerlaw regime followed by a sharp cut-off, like an exponential or Gaussian decay of the tail. single-scale networks: characterized by a connectivity distribution with a fast decaying tail, such as exponential or Gaussian Aging of the vertices: The vertex is still part of the network and contributing to network statistics, but it no longer receives links. The aging of the vertices thus limits the preferential attachment preventing a scale-free distribution of connectivities. Cost of adding links to the vertices or the limited capacity of a vertex: physical costs of adding links and limited capacity of a vertex will limit the number of possible links attaching to a given vertex.
In-degree Distribution
Out-degree Distribution
This broadcast method will find the target file quickly, given that it is located within a radius of ttl. However, broadcasting is extremely costly in terms of bandwidth. Such a search strategy does not scale well. As query traffic increases linearly with the size of GNUTELLA graph, nodes become overloaded.
Outline:
As the number of ISPs has grown, a new type of network access point, called a metropolitan area exchange (MAE) has arisen. There are about 50 such MAE around the U.S. today. Sometimes large regional and local ISPs also have access directly to NAPs.
ISP at the same level usually do not charge each other for exchanging messages. This is called peering. Higher level ISPs, however, charge lower level ones (national ISPs charge regional ISPs which in turn charge local ISPs) for carrying Internet traffic. Local ISPs, of course, charge individuals and corporate users for access.
Connecting to an ISP
ISPs provide access to the Internet through a Point of Presence (POP). Individual users access the POP through a dial-up line using the PPP protocol. The call connects the user to the ISPs modem pool, after which a remote access server (RAS) checks the userid and password. Once logged in, the user can send TCP/IP/[PPP] packets over the telephone line which are then sent out over the Internet through the ISPs POP.
ISP POP
ISP POP
Corporate T1 Customer T1 CSU/DSU Layer-2 Switch Corporate T3 Customer T3 CSU/DSU ATM Switch
ISP POP
NAP/MAE
Internet Organization
CN
NAP
POP
ISP
CN
CN
CN
POP
CN
NAP
POP
CN
POP
ISP
CN
CN
ISP = Internet Service Provider BSP = Backbone Service Provider NAP = Network Access Point POP = Point of Presence CN = Customer Network
Customer Network
Clients LAN
Ethernet 10 Mb/s
Servers
Router
WAN
T1 Link 1.54 Mb/s
NAP Architecture
ISP ISP ISP Backbone Operator
Routers
Backbone Operator
ISP
Tier 1 ISP
NAP
Tier 1 ISP
Tier 1 ISP
Tier-1 IP backbone
POP
The backbone is a set of POPs (usually one per city) Point-of-Presence (POP) : A collection of routers and switches housed in a single location
Tier-2 ISP pays tier-1 ISP for connectivity to rest of Internet tier-2 ISP is customer of tier-1 provider
Tier-2 ISP
Tier-2 ISP
Tier 1 ISP
NAP
Tier-2 ISPs also peer privately with each other, interconnect at NAP
Tier-2 ISP
Tier 1 ISP
Tier-2 ISP
Tier 1 ISP
Tier-2 ISP
Local and tier3 ISPs are customers of higher tier ISPs connecting them to rest of Internet
local ISP
local ISP
Tier 1 ISP
NAP
Tier 1 ISP
Tier 1 ISP
Tier-2 ISP local ISP
local ISP
Tier 1 ISP
NAP
Tier 1 ISP
Tier-2 ISP local local ISP ISP
Tier 1 ISP
Tier-2 ISP local ISP
Architecture of a POP
Backbone links Backbone Router
Backbone Router
Peering
Access Router
Access Router
Access Router
Access Router
ISPs
Corporate networks
Web Servers
Dial-up
ISP Architecture
Access Network Architecture
Modem
Internet Backbone
ISDN
ISDN service access links terminate at the ISP POP Digital signal. Due to signal strength limitations, ISDN subscribers must be within 18000 feet of the CO At the customers end, an ISDN adapter card is required.
DSL
Central Office
Modem
DSLAM Circuit Switch ISP POP Modem Pool Web Cache Router
Internet Backbone
DSL Access
DSL typically provisioned at 1.5Mbps from ISP to customer and at 128kbs in the reverse direction. DSL Access Multiplexer (DSLAM) at CO terminates DSL signals from hundreds of customers. The IP data is multiplexed into a single ATM connection by DSLAM and forwarded to the ISP POP
Dedicated Access
Leased lines from 56Kbs to 155Mbps. No multiplexing of other customers traffic. Can lead to higher operational cost. Lines terminate at routers in the POP.
Router
Router
ISP Router
The backbone of a large ISP is typically a WAN spread out across a large geographic area. Backbone routers connect the individual links composing the backbone .
ISP Backbone
For reasons of robustness and load management, multiple backbone routers can be located in the same geographic location and connected via a LAN.
We consider all of the backbone routers and the connecting LAN to be a backbone node.
These backbone nodes, whether they contain one or more routers, will serve as the points of connection from the outside world to the backbone.
Access Router
ISP Backbone
Dial-in POP (Downstream)
Customers, including smaller ISPs, enterprise, are connected to backbone nodes via access routers. Access routers gain their connectivity to the backbone, because they are on the same LAN as one or more backbone routers.
Remember, the backbone nodes contain backbone routers, as well as these access routers.
Any backbone entry point is known as a point of presence (POP). Modem entry points are known as dial-in POPs or dial-in hubs. Entry points for other types of networks are known as broadband POPs.
Access Router
ISP Backbone
Large dial-in POP (Downstream) Backbone Router
In practice, only the largest customers connect directly to access routers. Other customers are aggregated at broadband points of presence (broadband POPs). These are basically LANs. The customers connect to routers on these LANs, and then these LANs connect to the access nodes Additionally, some very large dial-in POPs do connect directly to backbone routers. These typically service very large corporate offices.
ISP Backbone
Peer ISP Gateway Router
Gateway routers, which are also connected via LANs to backbone routers, connect ISPs to each other. The router is known as a gateway router, if it connects a peer or upstream ISP. Downstream ISPs generally connect via an access router, or directly to a backbone Router. So, a gateway router leads to a peer or upstream provider, whereas an access router leads to a downstream network.
Measuring ISP Topologies with Rocketfuel[8]: Rocketfule internet topology mapping engine The goal is to obtain realistic, router-level maps of ISP networks. Important influence on: - The dynamics of routing protocols - The scalability of multicast - The efficacy of proposals for denial-of-service tracing and response - Other aspects of protocol performance (Internet path selection) Real topologies are not publicly available - Confidential
Mapping techniques
Three categories of mapping techniques:
Selecting Measurements
Directed probing
Path reduction
Alias Resolution
IP identifier
Selecting Measurements
Directed probing
To employ BGP tables to identify relevant traceroutes and prune the remainder To identify redundant traceroutes
Only one traceroutes needs to be taken when
Path reduction
two traceroutes enter and leave the ISP network at the same point
Alias resolution
Mercator method
Sending traceroute-like probe(to a highnumbered UDP port but with a TTL of 255) directly to the potentially aliased IP address
Requirement: routers need to be configured to
send the UDP port unreachable response with the address of the outgoing interface as the source address: Two aliases should respond with the same source
Alias method
Proposed methods by Spring etc.
IP identifier hints
IP identifier helps to identify a packet for reassembly after fragmentation IP identifier is commonly implemented using a counter that is incremented after sending a packet
Ally, a tool for alias resolution, sends a probe packet to the two potential aliases Port unreachable responses, including the IP identifiers x and y Ally sends a third packet to the address that responded first
References:
[1] Kenneth Calvert, Matthew Doar, Ellen Zegura, Modeling Internet Topology. [2]. Michalis Faloudsos, Petros Faloudsos, Christos Faloudsos, On Power-law Relationships of the Internet Topology [3]. Lada A. Adamic,1, Rajan M. Lukose,1, Amit R. Puniyani,2, and Bernardo A. Huberman1, Search in power-law networks. [4]. L. A. N. Amaral, A. Scala, M. Barthlmy, & H. E. Stanley, 1997, Classes of small-world networks. http://polymer.bu.edu/~amaral/Content_network.html [5]. Ellen Zegura, Kenneth Calvert, How to model an Internetwork [6]. Stefan Bornholdt, Holger Ebel, World Wide Web scaling exponent from Simons 1955 model [7]. S. Halabi and D. McPherson, Internet Routing Architectures, 2nd ed., Cisco Press, Indianapolis, 2000. [8]. Neil Spring Ratul Mahajan David Wetherall, Measuring ISP Topologies with Rocketfuel