CP3397 Network Design and Security Lecture 4 WAN design - Principles and practice

Preview:

Citation preview

CP3397 Network Design and SecurityLecture 4 WAN design - Principles and practice

WAN design

Wide area networks constructed from private circuits (leased lines) need careful design to optimise performance minimise costs provide adequate service allow redundancy for fault tolerance

Backbone and access networks

WANs can be split into two parts Backbone network linking main

centres Access networks linking endpoints to

nearest backbone node

Both aspects of the network need designDifferent rules apply to each type

Principles1. Good designs have many well-utilized components2. In voice network utilization should be high (to

reduce costs) - in a data network utilization should be low (to reduce delay)

3. Aim for 50% utilization on all links (to balance cost/delay)

4. Aim to have as few links as possible under 50% utilization

5. Use “natural traffic centres” - found from “weight” calculation

6. Balance need for shortest path, economy of scale from high speed links and utilization

7. Most design algorithms need repeated application to give best results

Some definitionsGraphsGraphs

A,B,C etc. are vertices(nodes)A,B,C etc. are vertices(nodes) (A,X), (X,Y) etc. are edges(A,X), (X,Y) etc. are edges P,Q,Z is a cycle (loop)P,Q,Z is a cycle (loop) Degree of a node is the number of edges at the nodeDegree of a node is the number of edges at the node

– Degree Y =3, degree C=1Degree Y =3, degree C=1

X

Y

Z

P Q

A

B

D

C

Trees

A tree is a connected simple graph with no cycles e.g.

XY

Z

P Q

A

B

D

C

Star

A tree is a star if only 1 node has degree >1

X

Y

Z

PQ

A

B

D

C

Chains

A chain is a tree with no nodes of degree >2

XY

Z

P Q

A

B

D

C

Weighted graphs

Each edge has a value (e.g. link speed, cost, etc.) Weight of the edge ei = w(ei)

To optimise a connected graph find the graph with the minimum weight

The Minimal Spanning Tree (MST)

Finding the MST

Two algorithms Prim and Kruskal Prim starts by selecting a node, adding the “least expensive edge” iterates until tree is built

Kruskal achieves the MST by starting with a graph and cutting out edges

Example MST

Use of MSTs

Small design problems - few nodesHighly reliable links with low “downtime” or network can tolerate unreliability

Nodes ‘v’ reliability As the number of nodes increases reliability decreases (exponentially!)

Shortest path trees

SPTs are when the path between each pair of nodes has the lowest weight Can be found using Dijkstra’s algorithm

See Cahn p67 and Kenyon p102

MSTs and SPTs will be different Prim and Dijkstra algorithms can be combined to give

MST or SPT using parameter alpha Ref: R. S. Cahn (1998) “Wide area network design”, Morgan

Kaufmann, ISBN 1-55860-458-8 and the Delite design tool, http://www.mkp.com/wand.htm and mirrored on the CP3397 homepage

Ref: T Kenyon (2002) “High-performance data network design” Digital Press, ISBN 1-55558-207-9

Access design

Each node on a backbone may have a number of local access pointsAccess networks route all traffic to the local backbone node.With n nodes there are nn-2 spanning trees!There are algorithms that reduce the possibilities - e.g. Esau-Williams, Sharma (Cahn Chapter 5)

Backbone design

Aim is to minimise the degree of the nodes (X) (i.e. number of connections at each) andminimise the number of hops (H) between all the nodesThese need balancing to produce a credible designFully-connected n-node mesh (H=1, X=n)Star (H=2, X=1 for all except central node)

Design algorithm

Mentor algorithm (one of many) relies on radius and weight to

determine backbone radius is proximity-based weight is traffic-related (in and out)

Calculates merit of a site based on distance from centre of network weight - i.e. traffic

Algorithm steps

1. Cluster sites within radius (Rparm) around those with largest merit

2. Select backbone centre (smallest weight x distance)

3. Build backbone tree4. Find the sequence of all pairs on the tree

starting with the outside5. Choose homes (sites between each pair)

Algorithm steps

6. Consider each pair once and add links if

utilization is too high otherwise traffic is sent via home node This adds links between non-adjacent

sites

Example

Detail diagrams

South East

Midlands

Input data

Coordinates of sitesCost of linksTraffic between sites in kbytesUser population at each siteParameters Rparm, wparm, slack

Network design

Parameters Rparm 0.4 Wparm 1.0 Slack 0.0 Utilization 0.5

Cost 757991

Alternative robust design

Parameters Rparm 0.4 Wparm

1.0 Slack 0.5 Utilization

0.5

Cost 766717

Detail diagrams

Midlands area detail

512K128K3x

128k

Summary

Design will depend on traffic, costs, andOther desirable parameters such as slack and utilizationDesign methods for Access and backbone are differentMany algorithms are availableReal networks will require careful cost minimisation

Recommended