Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
co ad
v
co co ad ad
vw vw vw vw
vw E vw E
vf
f F
d
vf wg
co co
vf wf vw v w
ad ad
vf wg vw
co ad
vf vw vw
c z c z
s t x v V
x x vw E f g d vw
x x z vw E f F F
x x z vw E f g
x z z
Martin
Grtschel
36
A Glance at some Instances
I
n
s
t
a
n
c
e
|
V
|
d
e
n
s
i
t
y
[
%
]
m
i
n
i
m
u
m
d
e
g
r
e
e
a
v
e
r
a
g
e
d
e
g
r
e
e
m
a
x
i
m
u
m
d
e
g
r
e
e
d
i
a
m
e
t
e
r
c
l
i
q
u
e
n
u
m
b
e
r
k 267 56,8 2 151,0 238 3 69
B-0-E-20 1876 13,7 40 257,7 779 5 81
f 2786 4,5 3 135,0 453 12 69
h 4240 5,9 11 249,0 561 10 130
E-Plus Project
Martin
Grtschel
37
Region Berlin - Dresden
2877
carriers
50 channels
Interference
reduction:
83.6%
Martin
Grtschel
38
Region Karlsruhe
2877
Carriers
75 channels
Interference
Reduction:
83.9 %
Martin
Grtschel
39
The UMTS Radio Interface
Completely new story
see talk by Andreas Eisenbltter on Monday
Martin
Grtschel
40
Contents
1. Telecommunication: The General Problem
2. The Problem Hierarchy: Cell Phones and Mathematics
3. The Problem Hierarchy: Network Components and Math
4. Network Design: Tasks to be solved
Addressing Special Issues:
5. Frequency Assignment
6. Locating the Nodes of a Network
7. Balancing the Load of Signaling Transfer Points
8. Integrated Topology, Capacity, and Routing
Optimization as well as Survivability Planning
9. Planning IP Networks
10. Optical Networks
11. Summary and Future
Martin
Grtschel
41
G-WiN Data
G-WiN = Gigabit-Wissenschafts-Netz of the DFN-Verein
Internet access of all German universities
and research institutions
Locations to be connected: 750
Data volume in summer 2000: 220 Terabytes/month
Expected data volume in 2004: 10.000 Terabytes/month
Clustering (to design a hierarchical network):
10 nodes in Level 1a 261 nodes eligible for
20 nodes in Level 1b Level 1
All other nodes in Level 2
Martin
Grtschel
42
G-WiN Problem
Select the 10 nodes of Level 1a.
Select the 20 nodes of Level 1b.
Each Level 1a node has to be linked to two Level 1b nodes.
Link every Level 2 node to one Level 1 node.
Design a Level 1a Network such that
Topology is survivable (2-node connected)
Edge capacities are sufficient (also in failure situations)
Shortest path routing (OSPF) leads to balanced
capacity use (objective in network update)
The whole network should be stable for the future.
The overall cost should be as low as possible.
Martin
Grtschel
43
Potential node locations for the
3-Level Network of the G-WIN
Red nodes are potential
level 1 nodes
Cost:
Connection between nodes
Capacity of the nodes
Blue nodes are all
remaining nodes
Martin
Grtschel
44
Demand distribution
The demand scales with the
height of each red line
Aim
Select backbone nodes and
connect all non-backbone nodes to
a backbone node
such that the
overall network cost is minimal
(access+backbone cost)
Martin
Grtschel
45
G-WiN Location Problem: Data
set of locations
set of potential Level 1a locations (subset of )
set of possible configurations at
location in Level 1a
For , and :
connection costs from to
tr
p
p
ip
i
V
Z V
K
p
i V p Z k K
w i p
d
=
=
=
e e e
=
= affic demand at location
capacity of location in configuration
costs at location in configuration
1 if location is connected to (else 0)
1 if configuration is used at loc
k
p
k
p
ip
k
p
i
c p k
w p k
x i p
z k
=
=
=
= ation (else 0) p
Martin
Grtschel
46
G-WiN Location/Clustering Problem
min
1
1
e e e e
+
=
s
=
=
p
k k
ip ip p p
p Z i V p Z k K
ip
p
k k
i ip p p
i k
k
p
k
k
p
p
w x w z
x
d x c z
z
z const
Each location i must be connected to a Level 1 node
Capacity at p must be large enough
Only one configuration at each Location 1 node
All variables are 0/1.
# of Level 1a nodes
Martin
Grtschel
47
Solution: Hierarchy & Backbone
Martin
Grtschel
48
G-WiN Location Problem:
Solution Statistics
The DFN problem leads to ~100.000 0/1-variables.
Typical computational experience:
Optimal solution via CPLEX in a few seconds!
A very related problem at Telekom Austria has
~300.000 0/1-variables plus some continuous variables
and capacity constraints.
Computational experience (before problem specific fine tuning):
10% gap after 6 h of CPLEX computation,
60% gap after simplification
(dropping certain capacities).
Martin
Grtschel
49
Contents
1. Telecommunication: The General Problem
2. The Problem Hierarchy: Cell Phones and Mathematics
3. The Problem Hierarchy: Network Components and Math
4. Network Design: Tasks to be solved
Addressing Special Issues:
5. Frequency Assignment
6. Locating the Nodes of a Network
7. Balancing the Load of Signaling Transfer Points
8. Integrated Topology, Capacity, and Routing
Optimization as well as Survivability Planning
9. Planning IP Networks
10. Optical Networks
11. Summary and Future
Martin
Grtschel
50
Re-Optimization of
Signaling Transfer Points
Telecommunication companies maintain a signaling
network (in adition to their communication transport
network). This is used for management tasks such as:
Basic call setup or tear down
Wireless roaming
Mobile subscriber authentication
Call forwarding
Number display
SMS messages
Etc.
Martin
Grtschel
51
Signaling Transfer Point (STP)
CCD
CCD
CCD
CCD CCD
CCD
CCD
CCD
CCD
Link-Sets
STP
Cluster
Cluster
Cluster
CCLK
CCD=Common Channel Distributors, CCLK=Common Channel Link Controllers
CCD=routing unit, CCLK=interface card
Martin
Grtschel
52
STP Problem description
Target
Assign each link to a CCD/CCLK
Constraints
At most 50% of the links in a linkset can be assigned to a
single cluster
Number of CCLKs in a cluster is restricted
Objective
Balance load of CCDs
Martin
Grtschel
53
STP Mathematical model
Data
C set of CCDs j
L set of links i
Di demand of link i
P set of link-sets
Q set of clusters
Lp subset of links in link-set p
Cq subset of CCDs in cluster q
cq #CCLKs in cluster q
Variables
{ }
0,1 , , e e e
ij
x i L j C
1 =
ij
x
if and only if
link i is assigned to CCD j
Martin
Grtschel
54
STP Mathematical model
{ }
2
min
1
,
0,1
p
p q
q
ij
j C
i ij
i L
i ij
i L
ij
i L j C
ij q
i L j C
ij
L
y z
x i L
D x y j C
D x z j C
x p P q Q
x c q Q
x
e
e
e
e e
e e
= e
s e
> e
(
s e e
(
(
s e
e
= e
s e
> e
(
s e e
(
(
s e
e
s
Assign each link
Upper bound of CCD-load
Diversification
Lower bound of CCD-load
CCLK-bound
Integrality
Restricted number of changes!
Min load difference
Martin
Grtschel
59
STP Alternative Model
{ }
*
, ( )
2
min
1
,
0,1
p
p q
q
ij
i L j C j j i
ij
j C
i ij
i L
i ij
i L
ij
i L j C
ij q
i L j C
ij
L
x
x i L
D x y j C
D x z j C
x p P q Q
x c q Q
x
y z D
e e =
e
e
e
e e
e e
= e
s e
> e
(
s e e
(
(
s e
e
s
Capacity constraints : e e E
0 1
1 0
e
T
e e e
x x x = > > > >
0
e
T
t t
e e e
t
y c x
=
=
Martin
Grtschel
72
Model: Routings
Path variables :
Capacity constraints :
Demand constraints :
, ,
s
s uv
s S uv D P e e eP
( ) 0
s
uv
f P >
0
0
:
( )
uv
e uv
uv D
P e P
y f P
e
eP e
>
e E e
0
0
( )
uv
uv uv
P
d f P
eP
=
uv D e
Path length restriction
Martin
Grtschel
73
Model: Survivability (one example)
Path restoration reroute affected demands
H B
D
F
M
120
60
60
H B
D
F
M
for all seS, uveDs
for all seS, eeEs
60
60
H B
D
F
M
Martin
Grtschel
74
Mathematical Model
topology decisison
capacity decisions
normal operation routing
component failure routing
s
uv s
P D uv S s P e e e , , 0 ) ( > P f
s
uv
) (
:
0
0
P f y
D uv P e P
uv e
uv
e e P e
> E ee
P e
=
0
) (
0
uv
P
uv uv
P f d D uv e
1
min
e
T
t t
e e
e E t
k x
e =
e
T t E e , , 1 , = e
, 1, ,
e
e E t T e =
{0,1}
t
e
x e
t
e
t
e
x x >
1
=
=
e
T
t
t
e
t
e e
x c y
0
E ee
LP-based Methods
Feasible
integer
solutions
LP-based
relaxation
Convex
hull
Objective
function
Cutting
planes
Feasible
integer
solutions
LP-based
relaxation
Convex
hull
Objective
function
Cutting
planes
Flow chart
Optimal
solution
Initialize
LP-relaxation
Run
heuristics
Solve
LP-relaxation
Separation
algorithms
Augment
LP-relaxation
Solve feasibility
problem
Separation
algorithms
Inequalities?
No
Yes
x variables
integer?
Yes
No
Feasible
routings?
Yes
No
Polyhedral combinatorics
Valid inequalities (facets)
Separation algorithms
Heuristics
Feasibility of a capacity vector
LP-based approach:
Martin
Grtschel
77
Finding a Feasible Solution?
Heuristics
Local search
Simulated Annealing
Genetic algorithms
...
Nodes Edges Demands Routing-Paths
15 46 78 > 150 x 10e6
36 107 79 > 500 x 10e9
36 123 123 > 2 x 10e12
Manipulation of
Routings
Topology
Capacities
Problem Sizes
Martin
Grtschel
78
How much to save?
Real scenario
163 nodes
227 edges
561 demands
34% potential savings!
==
> hundred million dollars
PhD Thesis:
http://www.zib.de/wessaely
wessaely@atesio.de
Martin
Grtschel
79
Contents
1. Telecommunication: The General Problem
2. The Problem Hierarchy: Cell Phones and Mathematics
3. The Problem Hierarchy: Network Components and Math
4. Network Design: Tasks to be solved
Addressing Special Issues:
5. Frequency Assignment
6. Locating the Nodes of a Network
7. Balancing the Load of Signaling Transfer Points
8. Integrated Topology, Capacity, and Routing
Optimization as well as Survivability Planning
9. Planning IP Networks
10. Optical Networks
11. Summary and Future
Martin
Grtschel
80
Comment
9. Planning IP Networks
10. Optical Networks
11. Summary and Future
The lecture ended after about 100 minutes. The last
three topics above were not covered.
Martin
Grtschel
81
Summary
Telecommunication Problems such as
Frequency Assignment
Locating the Nodes of a Network Optimally
Balancing the Load of Signaling Transfer Points
Integrated Topology, Capacity, and Routing Optimization
as well as Survivability Planning
Planning IP Networks
Optical Network Design
and many others
can be succesfully attacked with optimization techniques.
Martin
Grtschel
82
Summary
The mathematical programming approach
Helps understanding the problems arising
Makes much faster and more reliable planning possible
Allows considering variations and scenario analysis
Allows the comparision of different technologies
Yields feasible solutions
Produces much cheaper solutions than traditional
planning techniques
Helps evaluating the quality of a network.
There is still a lot to be done, e.g.,
for the really important problems,
optimal solutions are way out of reach!
Martin
Grtschel
83
The Mathematical Challenges
Finding the right ballance between
flexibility and controlability of future networks
Controlling such a flexible network
Handling the huge complexity
Integrating new services easily
Guaranteeing quality
Finding appropriate Mathematical Models
Finding appropriate solution techniques (exact,
approximate , interactive, quality guaranteed)
Martin Grtschel Institute of Mathematics, Technische Universitt Berlin (TUB)
DFG-Research Center Mathematics for key technologies (FZT 86)
Konrad-Zuse-Zentrum fr Informationstechnik Berlin (ZIB)
groetschel@zib.de http://www.zib.de/groetschel
Mathematical Challenges
in Telecommunication
The End
Martin Grtschel
IMA workshop on
Network Management and Design
Minneapolis, MN, April 6, 2003