Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
David Tipper
Associate Professor
Department of Information Science and
Telecommunications
University of Pittsburgh
Slides 8
http://www.sis.pitt.edu/~dtipper/2110.html
EIR
• Can roughly categorize access
AUC
design problems
HLR IBM
VLR
– One speed one center design
MSC • Capacitated MST problem
*8
P
Ba
yNetwok
x5
OO1
AON
6
0 RST
30
rs
WR
P
ET
HER LINK
LM
A
A
RS23
ALM
FN0FAN
1 P
2C
WR0PWR1
INS ACT AL
M
PCC
ARD
Centiil on1400
D
S
*8
P
BayNe
x5
OO1
AON
6
0
30
t wok
RS
T
rs
ETH
ER IK
L N R
S232
C INS AC
T LM
A
CCARD
P
Centil il on1400
D
S
– Esau-Williams algorithm
– Sharma algorithm
LM
A
PWR AL
M
FAN
0 F
AN1PWR
0 P
WR1
BSC
BSC
D
S
Ba
yNetwok
rs Centiil on1400
*8
P x5
0 ET
HER LINK RS23
2C INS ACT AL
M
RST
OO130
AON
6
PCC
ARD
• Examples
ALM
WR
P LM
A
F
AN0FAN
1 P
WR0PWR1
BSC
BS1
BS4
BS7
BS1
– Multi-speed access design
BS5
BS7
BS5 BS6
• Have multiple link speeds/types
BS6 – For example, LAN design from variety of
hosts
– Multi-center Design
• Cellular networks with multiple base
station controllers
TELCOM 2110 2
1
Access Network Design
• For Multi-speed and Multi-Center network design
need
• Traffic Matrix
– Specifies traffic between all source-destination pairs
– Entry Tij is the traffic from source i to destination j
– Source and destination maybe host, LAN, etc.
– Developed in conceptual design- usually based on
peak busy hour
• Cost Matrix
– Cost of link capacity Cij between nodes i and j
– Cost may depend on traffic demand w needed C,ij(w)
• Nodal Weight
– The total traffic demand at a node
TELCOM 2110 3
– weight(Ni) = Σj (Ti,j+Tj,i).
– Weight of a site is the sum of all traffic
Dallas ----------- 2.1 Mb
in and out of the site .1Mb
– Link Capacity needed is proportional
to weight
• Example of corporation in Dallas,
Vienna, and Denver Denver .3 Mb ----------- .8Mb
----
Weight(Dallas) = 2.2 + 2.8 = 5.0 Mbps
Weight(Denver) = 1.1 + 1.6 = 2.7 Mpbs
Weight(Vienna) = 4.0+ 2.9 = 6.9 Mbps
Vienna 2.5 Mb 1.5 Mb ------------
TELCOM 2110 4
2
Backbone & Access Sites
• Design Principle
– Compute the weight of all the nodes to
determine if there are natural traffic centers
if the network is flat (i.e., no facility hierarchy
– Example of corporation - Vienna has largest weight
• Generally acceptable for small weight nodes to
route their traffic via big weight nodes, but we
do not want to route the traffic between big
nodes via the small nodes
• Not always obvious what is big and small node
– Want large difference in weight between the big and
small
TELCOM 2110 5
3 4 5
6
TELCOM 2110 6
3
Multiple Speed Access Design
• Formulation/algorithm based on concept of predecessor and ancestor
• A tree T rooted at a node Root can be represented uniquely by a
predecessor function pred
– The function pred( ) gives the node one hop closer to the root from the
node in question
– pred(6) = 3, pred(3) = 1, pred(1) = root, pred(4) = 2, etc.
– pred2(6) = pred(pred(6)) = 1, pred3(6) = root
• Define pred(root) = root
• Given a tree T and the associated predecessor function, the ancestors
of N are all the nodes N’ that are downstream from the node away
from the root node.
Root
Ancestor(1) = {3,6}
Ancestor(2) = {4,5} 1 2
Ancestor(3) = {6}
(bit of a misnomer !)
3 4 5
pred n (N’) = N for some n > 0
6
TELCOM 2110 7
• Given
– set of nodes N0, N1, N2, … , Nn.
– set of weights (traffic demand) w1, w2, … , wn for each node
– set of link types L1, L2, … , Lm
– Set of capacities W1, W2, … , Wm
– cost matrix C(i, j, k) that gives the cost of a link of type Lk between Ni
and Nj
• Find: the tree rooted at N0 and the link assignments such that
∑ C (i , j , l )
(i)
Minimize
l∈Links
(ii)
∑w (i ) < WLink ( N , pred ( N ))
i∈N ∪ Ancestors ( N )
TELCOM 2110 8
4
Multiple Speed CMST Problem
• Consider Multi-speed capacitated MST problem constraint
∑w (i )
i∈N ∪ Ancestors ( N )
< WLink ( N , pred ( N ))
Root
For example link(2,0)
0
w2 + w4 + w5 < W(2,0)
1 2
For link(1,0)
w1 + w3 + w6 < W(1,0)
TELCOM 2110 10
5
MSLA Example
Consider network with four access nodes to connect to a hub want a
MAXIMUM link utilization of 50%
Link cost are a piecewise linear function of distance and data rate
TELCOM 2110 11
MLSA Example
• Link Costs
TELCOM 2110 12
6
MSLA Example
L1
spare_capacity(1)=0.5*56000-
20000=8000
L1
spare_capacity(2)=0.5*9600-
2400=2400
spare_capacity(3)=0.5*56000-
9600=18400
spare_capacity(4)=0.5*9600-4800=0
TELCOM 2110 13
MSLA Example
= 10 + (15-7) -10 = 8 L1
Tradeoff1(3)= C(1,3, l) +
Upgrade(3,20000) – C(1,0, 1 ) L1
= 12 + (20-10) -10 = 12
Tradeoff1(4)= C(1,4, l) +
Upgrade(4,20000) – C(1,0, 1 )
= 12 + (15 -7) -10 = 10
Tradeoff(1) = min{8,12,10} = 8
TELCOM 2110 14
7
MSLA Example
Tradeoff2(3)= C(2,3, l) + L0
Upgrade(3,2400) – C(2,0, 1 )
L1
= 5 + 0 -7 = -2
Tradeoff2(4)= C(2,4, l) + L1
Upgrade(4,2400) – C(2,0, 1 )
= 6 + (15 -7) -7 = 7
Tradeoff(2) = min{0,-2,7} = -2
TELCOM 2110 15
MSLA Example
Tradeoff3(2)= C(3,2, l) + L0
Upgrade(2,9600) – C(3,0, 1 )
L1
= 12 + (15-7) -10= 10
Tradeoff3(4)= C(3,4, l) + L1
Upgrade(4,9600) – C(3,0, 1 )
= 10 + (15 -7) -10 = 8
Tradeoff(3) = min{12,10,8} = 8
TELCOM 2110 16
8
MSLA Example
MSLA Example
L0 Iteration 1 topology
L1
L1
Iteration 2: Note adjust node weights
L0
Repeat tradeoff calculations
L0
Result is Let N4 goes through N3,no
Iteration 2 upgrade is needed and it is the best
tradeoff value
L1
Iteration 2 topology
L1
TELCOM 2110 18
9
MSLA Example
Iteration 2
L0
L0
L1
L1
L0
L0 • Iteration 3 results in connecting N3 to N1
Final
design and increase (1,0) to 256 Kbps link.
L1
• All tradeoff values are positive - STOP
L2
TELCOM 2110 19
TELCOM 2110 20
10
Esau Williams: 20 nodes with 9.6Kbps links
Cost = $26,963
Only 9 sites share links to N0, more like a star.
TELCOM 2110 21
Cost = $30,160
A nice tree structure, but the cost is higher because out on
the periphery of the network there is too much capacity.
TELCOM 2110 22
11
MSLA: 20 nodes with multispeed links
Cost = $22,760
There is a central D56 tree and a peripheral D96 tree
TELCOM 2110 23
x5
OO1
AON
6
0 RST
30
rs
ET
HER LINK RS23
2C INS ACT AL
M
PCC
ARD
Centiil on1400
D
S
MSC
– One speed one center design
D
S
BayNe
t wok
rs Centil il on1400
ALM
WR
P LM
A
A
FN0FAN
1 P
WR0PWR1
CCARD
P
LM
A
PWR AL
M
FAN
0 F
AN1PW0
R P
WR1
BSC
BSC
D
S
Ba
yNetwok
rs Centiil on1400
*8
P x5
0 ET
HER LINK RS23
2C INS ACT AL
M
RST
OO130
AON
6
PCC
ARD
ALM
WR
P LM
A
A
FN0FAN
1 P
WR0PWR1
BSC
– Esau-Williams algorithm
BS3
– Sharma algorithm
BS3 BS2 BS4
BS2
BS1
BS4
BS7
BS1
BS5
– Multi-speed access design
BS7
BS6
BS5 BS6
• mCMST – MSLA
– Multi-center Design
• Multiple Centers (hubs) – nodes
can connect to any center
TELCOM 2110 24
12
MultiCenter Access Design
• Given
– A set B of hub or backbone sites {B0, …, Bm}
– A set N of access nodes {N1, … , Nn}
• A set of weights {w1, … , wn} for each access node
• A upper limit on capacity, W (one speed design).
• A cost matrix Cost(i,j) giving the costs between each (hub)
backbone/access pair of sites.
• Build a set of trees that connect each access node to a hub
• Constructing a forest of trees – often not interconnected
TELCOM 2110 25
∑ N i ∈T j
wi < W
∑ ∑Trees l∈Links
C ( Node1l , Node2l )
TELCOM 2110 26
13
Example
Consider site with 3 backbone nodes
Circles : X, Y and Z
TELCOM 2110 27
14
Nearest-Neighbor Esau-Williams (NNEW)
TELCOM 2110 29
TELCOM 2110 30
15
Nearest-Neighbor Esau-Williams (NNEW)
TELCOM 2110 31
TELCOM 2110 32
16
Nearest-Neighbor Esau-Williams (NNEW)
TELCOM 2110 33
• MCEW has an advantage over NNEW when size of nodes in clusters large
TELCOM 2110 34
17
NNEW vs. MCEW
TELCOM 2110 35
18
Some Branches Have Too Many Nodes.
TELCOM 2110 37
19
Some Node in Tree Has Too Many Links
TELCOM 2110 39
20