LAN/WAN Optimization Techniques

1
LAN/WAN Optimization Techniques

• Harrell J. Van Norman
• Presented by Lin Shu-Ping

2
Outline

• Design Tools as Part of the Total Network
  Engineering Process
• Network Design Tool Utilization
• Network Design Tool

3
Design Tools as Part of the Total Network
Engineering Process

• NDTs make up only one step in the total network
  engineering process.

4
Prior To Use

• Requires collecting current traffic measures and
  forecasting for network growth.
• The precision of any NDT is related to
  correctness of the design criteria upon which
  analysis is based.
• Typical questions asked by NDT
• Message profiles of each applications
• Protocol characteristics
• Environment of transmission network
• Traffic Profiles

5
Prior To Use

• When using NDT accurately, defining design
  criteria is crucial.
• Integration of NDT into a network management
  system is a more credible approach.
• The optimization of the network will be
  proportional to precision of the design criteria.

6
During Use

• The network engineer postulates with what if
  scenario by changing node location.
• Using NDT involves iterative refinement generally
  produced during evaluating of various network
  alternatives.

7
Subsequent To Use

• Additional phases of network engineering include
  reconfiguration, equipment acquisition,
  verification, installation and administrator.
• Iterative improvements are made by evaluating
  network costs and performance against the
  operational criteria.

8
Outline

• Design Tools as Part of the Total Network
  Engineering Process
• Network Design Tool Utilization
• Network Design Tool

9
Network Design Tool Utilization

• NDT utilization step includes
• Selecting a design technique
• Acquiring a tool
• Developing a model
• Analyzing the model

10
Design Technique Selection

• Two basic design techniques
• Discrete event simulation
• Analytic heuristic modeling
• If network engineer is interested only in network
  performance, using simulation.
• If the network engineer needs circuit cost and
  network performance combined, analytic heuristic
  is the preferred approach.

11
Tool Acquisition

• This step involves acquiring an NDT from among
  many suppliers of NDTs.
• You get what you pay for is generally true
  within the NDT market.

12
Model Development

• Generate a logical model of the network that
  requires analysis and design.
• The model is based on a set of locations for data
  input and termination.

13
Model Analysis

• Test model with various inputs and observe the
  resulting cost and performance outputs.
• Developing allowable ranges for acceptable input
  parameters assists in insuring the model accuracy.

14
Outline

• Design Tools as Part of the Total Network
  Engineering Process
• Network Design Tool Utilization
• Network Design Tool

15
Network Design Tool

• Technical approach
• Analytic heuristic
• Discrete event simulation
• Design algorithms
• Multipoint line connection
• Backbone design
• Topological structures
• Tree, Ring, Star, String

16
Technical approach

• There is nearly overwhelming number of
  possibilities in configuring a network.
• Insuperable amounts of computing times would be
  required for applying algorithms to find optimum
  constrained design.
• It is appropriate to use heuristic techniques or
  simulation-based approaches.

17
Heuristics

• Heuristics are chosen approximations of actual
  analytic calculations.
• Using heuristics is necessary whenever
  computational time and resources would be
  excessive to provide actual analytic solution.

18
Heuristics (cont.)

• No NDT can produce totally optimal design due to
  inaccurate input values.
• Evidence indicate that good heuristic algorithms
  can produce network designs that are within 5
  percent of optimal solution.

19
Simulation

• Network simulation predicts performance
  characteristics.
• Whenever extremely precise performance evaluation
  is necessary, simulation is the preferred
  technique.
• The results of simulation predict how networks
  will perform under various loads.

20
Simulation (cont.)

• Simulations overcome deficiencies inherent in
  entirely analytic heuristic algorithm for
  predicting network reliability.

21
Design Algorithms

• When developing a network design, there are two
  basic classes of problems
• Developing acceptable line loading
• Optimal line configurations
• All NDTs address line loading constraints, with
  simulation models providing precise estimation of
  end-user response time.

22
Design Algorithms (cont.)

• Regardless of the method from determining
  acceptable lineloading constraints,designing a
  network configuration is necessary.

23
Multipoint Line Connection

• Multipoint lines reduce total circuit mileage
  costs by enabling multiple users to share
  circuits.
• To minimize the cost of that line involves
  computing the minimal spanning tree.
• Minimal spanning tree calculations are exact
  optimal algorithms with link-loading constraints
  or unconstrained limits.

24
Esau-Williams Algorithm

• Start with the simplest type of network
• One with a central controller hub connected to
  each remote terminal by a separate circuit.
• Such network can be accepted when terminal are
  very heavily loaded or equipment precludes line
  sharing.
• Set aside each fully loaded line, because it
  obviously cannot be multipointed.

25
Esau-Williams Algorithm (cont.)

• In each iteration the node with the greatest
  differential distance from the hub to the nearest
  neighboring node is located.
• It reconnects that node to its nearest
  neighboring node, thereby providing the greatest
  cost benefits.
• In this manner, each iteration removes one
  expensive link and replaces it with the best
  alternative link.

26
Esau-Williams Algorithm (cont.)
27
Prim Algorithm

• Prims algorithm functions in the reverse of the
  Esau-Williams algorithm.
• It selects the nodes closest to the center then
  connects in those node that are closest to those
  already in the network.
• Minimizing the maximum costs by means of the
  Esau-Williams algorithm yields improved designs
  over Prims algorithm.

28
Prim Algorithm (cont.)
29
Concentrator Placement

• Given potential concentrator sites,determine the
  number and locations of concentrators and assign
  each terminal to concentrator.
• Add and Drop Algorithms

30
Add and Drop Algorithms

• Step1Clustering nearby nodes into COM nodes,
  thereby reducing the problem in size and
  converting to a point to point formulation.
• Step2Using the add algorithm to partition the
  COM nodes with other COM nodes that give the
  greatest cost benefits by being connected to
  generic access facilities instead of resource
  connection point.

31
Add and Drop Algorithms (cont.)

• Step3Local optimization by selecting one
  specific node site for the generic access
  facility.
• Step4Line layout by replacing all the COM nodes
  with the actual nodes.

32
Backbone Design

• In hierarchical network involves two design
  problems.
• Design of the regional subnetworks
• Backbone portion of the network
• The cut saturation algorithm is a common example
  of backbone design algorithm.
• It iteratively finds the least-cost backbone
  network for a specified throughput, subject to
  time delay and reliability constraints.

33
Cut Saturation Algorithm

• Cut saturation algorithm consists of five basic
  steps in any one iteration
• Routing
• Saturated cutset determination
• Add-only step
• Delete-only operation
• Perturbation setp

34
Cut Saturation Algorithm (cont.)

• Routing
• setting up at each node along the path a routing
  table directing messages with a particular
  destination address to appropriate outgoing link.
• Saturated cutset determination
• Links are ordered according to their utilization
• Links are then removed, one at a time, in order
  of utilization
• The minimal set that disconnects the network is
  called a saturated cutset

35
Cut Saturation Algorithm (cont.)

• Add-only step
• Adding the least-cost links to the network that
  will direct traffic from the saturated cutset.
• Nodes that are at least two links removed from
  the cutset are chosen as candidates for possible
  linkages.
• Delete-only step
• Links from a highly connected topology are
  eliminated.
• One link at a time is removed at each iteration
  by finding maximum cost link.

36
Cut Saturation Algorithm (cont.)

• Perturbation step
• Once a desired throughput range has been
  attained, network links are rearranged by
  add-only and delete-only operations to reduce
  cost.
• Add-only and delete-only operations are used
  sequentially as long as throughput remains within
  the bounds.

37
Routing and Service Options

• Effective design dynamics include
• Fractional
• Hubless
• LEC Bridging
• Routing Strategies

38
Routing and Service Options (cont.)

• Three service options
• Total service
• Coordinated service
• Baseline service

39
Routing and Service Options (cont.)

• Total service
• ATT will design, order, and bill the entire
  circuit.
• Require the highest degree of dependence upon
  ATT
• Coordinated service
• Carrier responsibility for ordering and
  maintaining the circuit as well as a measure of
  customer control over the network

40
Routing and Service Options (cont.)

• Baseline service
• Taking all responsibilities for their network
• Sophisticated diagnostic equipment and
  experienced technicians should be on hand

41
Topologies Supported

• Topology of a network may be organized according
  to two level
• Terminal access network (TAN)
• Backbone-mesh network (BMN)
• For BMN, satisfactory design are typically star,
  ring, and hyper-ring.
• NDT should support these network topologies.

42
Structures Evaluated

• Average number of links per node
• A higher link-per-node ration indicates a more
  expensive network topology
• Maximum number of intermediate nodes
• The accessibility of any node to any other node
• A large number of intermediate node results in
  higher delay

43
Structures Evaluated (cont.)

• Maximum node capacity
• A measure of node vulnerability, defined as the
  maximum number of links that connect to a given
  node.
• Amount of traffic a specific node is required to
  support.
• Number of nonredundant routes
• A measure of network reliability
• A high number of nonredundant routes points to a
  network topology with good reliability.

44
Structures Evaluated (cont.)

• Total interconnect have a high cost and degree of
  reliability
• Tree, star, and string have a low cost and degree
  of reliability
• Ring and hyper-ring have relatively low cost
  combined with high degree of reliability.

45
Tariff

• Tariff are descriptions of telecom services and
  prices of those services.
• Accurate and completer tariff data is essential
  for bill verification.
• Telecom companies provide three basic types of
  transmission services private line, switched,
  and packet services.

46
Tariff (cont.)

• Three methods of obtaining tariff data
• Getting data directly from tariff database
  supplier requires the least cost by the NDT
  provider
• Obtaining from tariff data supplier and then
  incorporated into internal database structures
  requires significantly greater effort.
• Obtaining tariff data directly from the FCC or by
  subscribing to the filing bodies themselves
  demands the greatest degree of effort and skill
  by NDT provider.