2-Ad Hoc Networks-Graph Theory-Lecture2

Ad Hoc Networks:#
Architectures & Protocols
Dr Ljiljana Simić
iNETS, RWTH Aachen University
SS2016

Important Note:
These course notes may contain some copyrighted
material. The copyright of this material covers its
use in class and for educa7onal purposes, but you
are not allowed to distribute this course material
freely. Under the code of appropriate use, please
refrain from uploading the provided source files or
documents to any publicly accessible system outside
RWTH Aachen University without prior permission.
1
– Graph Theory &#
Routing Fundamentals
#

Some Applications of Graph Theory

§  models for communica7ons and electrical networks
§  models for computer architectures
§  network op7miza7on models for opera7ons analysis,
including scheduling and job assignment
§  analysis of Finite State Machines
§  parsing and code op7miza7on in compilers
ver.ces
edges
2
Graphs ↔ Networks
Graph Ver7ces Edges
Flow
(Network) (Nodes) (Arcs)
telephone exchanges, cables, fiber op.cs,
voice, video,
communica.ons computers, satellites, microwave/wireless
data packets
radios links
gates, registers, processors
circuits wires current
mechanical joints rods, beams, springs heat, energy
reservoirs, pumping
hydraulic sta7ons, lakes
pipelines fluid, oil
financial stocks, currency transac7ons money
freight,
airports, rail yards, street highways, railbeds,
transporta7on intersec7ons airway routes
vehicles,
passengers
Application to Ad Hoc Networking

§  networks can be represented by graphs
§  the mobile nodes are ver.ces
§  the wireless communica7on links are edges
ver.ces
edges
§  rou7ng protocols oNen use shortest path algorithms

§  this lecture is background material for ad hoc rou.ng
algorithms
3
Elementary concepts
ver.ces
§  a graph G(V,E) is two sets of objects
-‐ ver.ces (or nodes), set V
-‐  edges, set E
edges
§  a graph is represented with dots or circles (ver7ces),
joined by lines (edges)

§  the magnitude of graph G is characterized by:

-‐  number of ver7ces |V| (called the order of G) and
-‐  number of edges |E| (size of G)

§  running 7me of algorithms is measured in terms of the
order & size
Directed graph
§  an edge e∈ E of a directed graph is represented as an
ordered pair (u,v), where:
à u, v ∈ V
à u is the ini7al vertex and v is the terminal vertex
à assume that u ≠ v
2
3
1
V = { 1, 2, 3, 4}, |V| = 4

4
E = {(1,2), (2,3), (2,4), (4,1), (4,2)}, |E|=5
4
Undirected graph
§  an edge e∈ E of an undirected graph is represented as an
unordered pair (u,v)=(v,u), where:
à u, v ∈ V
à assume that u ≠ v
2
3
1
V = { 1, 2, 3, 4}, |V| = 4

4
E = {(1,2), (2,3), (2,4), (4,1)}, |E|=4
Degree of a vertex
§  in an undirected graph,
degree of a vertex is the number of edges incident on it
§  in a directed graph,
à out degree of a vertex is the number of edges leaving it
à in degree of a vertex is the number of edges entering it
2 2
3 3
1 1
4 4
The degree of vertex 4 is 2. The out degree of vertex 4 is 2,
the in degree of vertex 4 is 1.
5
Weighted graph
§  a weighted graph is a graph for which each edge has an
associated weight,
usually given by a weight func4on, w: E → R
2 2
1.2 0.5 4 2
6
0.2 3 3
1 1
9
2.1 8
4 4
Walks and paths
3
V2 V3
1
2
3
V1 1
V6
4
4
V4 V5
1
6
Walks and paths
§  a walk is an sequence of nodes (v1, v2,..., vL) such that
{(v1, v2), (v2, v3),..., (vL-‐1, vL)} ⊆
E, e.g. (V2, V3, V6, V5, V3)

3
V2 V3
1
2
3
V1 1
V6
4
4
V4 V5
1
Walks and paths

§  a simple path is a walk with no repeated nodes,
e.g. (V1, V4, V5, V2, V3)
3
V2 V3
1
2
3
V1 1
V6
4
4
V4 V5
1
7
Walks and paths

§  a cycle is a walk (v1, v2,..., vL) where v1=vL, with no other nodes
repeated and L>3, e.g. (V1, V2, V5, V4, V1)
§  a graph is called cyclic if it contains a cycle;

otherwise it is called acyclic

3
V2 V3
1
2
3
V1 1
V6
4
4
V4 V5
1
Complete graphs
§  a complete graph is an undirected/directed graph in which
every pair of ver7ces is adjacent
§  if (u, v) is an edge in a graph G,

we say that vertex v is adjacent to vertex u.

A B A
B
C
D C
§  4 nodes and (4*3)/2 edges §  3 nodes and 3*2 edges
§  V nodes and V*(V-‐1)/2 edges §  V nodes and V*(V-‐1) edges
8
Connected graphs
§  an undirected graph is connected if you A B C
can get from any node to any other by
following a sequence of edges OR any
two nodes are connected by a path
D E F
§  a directed graph is strongly connected if A B
there is a directed path from any node

to any other node
C D
§  a graph is sparse if |E| ≈ |V|

§  a graph is dense if |E| ≈ |V|2
Bipartite graph
§  a bipar4te graph is an undirected graph G=(V,E)
in which V can be par77oned into 2 sets V1 and V2,
such that (u,v) ∈ E implies either:
u ∈ V1 and v ∈ V2
or V1 V2
v ∈ V1 and u ∈ V2.
u1
v1
u2
v2
u3
v3
u4
9
Trees
Trees
Let G = (V, E ) be an undirected graph.
à the following statements are equivalent: A

1.  G is a tree

2.  any two ver.ces in G are connected by a
B C
unique simple path
3.  G is connected, but if any edge is removed
from E, the resul7ng graph is disconnected
4.  G is connected, and |E| = |V| -‐1 D E
5.  G is acyclic, but if any edge is added to E,
the resul7ng graph contains a cycle
6.  G is acyclic, and |E| = |V| -‐1 F

10
Spanning Tree
§  a tree (T) is said to span G = (V,E) if T = (V,E’) and E’ ⊆ E
§  for a given graph there are usually V2 V3

several possible spanning trees, e.g.:

V2 V3 V1
V6
V6 V4 V5
V1
T1
V2 V3
G V4 V5
V6
V1
T2 V4
V5
Applications of Trees
§  unicast rou7ng (one-‐to-‐one) → shortest path spanning tree
§  mul7cast rou7ng (one-‐to-‐several)
§  maximum probability of reliable one-‐to-‐all communica7ons
→ maximum-‐weight spanning tree
§  load balancing → degree-‐constrained spanning tree
11
Minimum Spanning Tree
§  given connected graph G with real-‐valued edge weights ce,
a Minimum Spanning Tree (MST) is a spanning tree of G
whose sum of edge weights is minimized
3
2 24 3 2
4 4
1 1
23 9 9
6 18 6
6 6
5 4 5 4
16 11 11
8 5 8 5
7 7
10 14
7 21 8 7 8
G = (V, E) T = (V, F) w(T) = 50

Cayley's Formula (1889)
There are nn-‐2 spanning trees of a complete graph Kn
§  n = |V|, m = |E|
§  can't solve MST by brute force (because of nn-‐2)
Applications of MST
MST is central combinatorial problem with diverse applica.ons
§  designing physical networks
§ telephone, electrical, hydraulic, TV cable, computer, road

§  cluster analysis

§ delete long edges, leaves connected components
§ finding clusters of quasars and Seyfert galaxies
§ analyzing fungal spore spa7al pakerns

§  approximate solu7ons to NP-‐hard problems

§ metric TSP (Traveling Salesman Problem), Steiner tree

§  indirect applica7ons

§ describing arrangements of nuclei in skin cells for cancer research
§ learning salient features for real-‐7me face verifica7on
§ modeling locality of par7cle interac7ons in turbulent fluid flow
§ reducing data storage in sequencing amino acids in a protein

12
MST computation
Prim’s Algorithm
§ select an arbitrary node as the ini.al tree (T)
§ augment T in an itera.ve fashion by adding the outgoing edge
(u,v) (i.e. u ∈ T and v ∈ G-‐T ) with minimum cost (i.e. weight)
§  the algorithm stops aNer |V| -‐ 1 itera7ons
§  computa7onal complexity = O (|V|2)

Kruskal’s Algorithm
§ select the edge e ∈ E of minimum weight → E’ = {e}
§ con7nue to add the edge e ∈ E – E’ of minimum weight that
when added to E’, does not form a cycle
§  computa7onal complexity = O (|E|log|E|)
Prim’s Algorithm (example)

V2 3 V3 V1
V2 V3
1 Algorithm starts
2 3
3 V6 V2
V1 1 1
1
4
1 4 V1
V1
V4 V5
ANer the 1st itera7on ANer the 2nd itera7on
V3 V3
V2 3 V2 3 V2 3 V3
1 1 1
2
1 1 1
V1 V1
V1 V6
1 1
V5 V4 V5 V4 V5
ANer the 3rd itera7on ANer the 4th itera7on ANer the 5th itera7on
13
Kruskal’s Algorithm (example)
V2 3 V2 V3
V3 V2
1
1 1
2 V1 Algorithm
3 V6 starts 1
V1 1 V1
V2 V3 V4
4 1
1 4
V4 V5 V1 1 V5
ANer the 1st 1
itera7on ANer the 2nd itera7on
V5
V2 V3 V2 3 V3
1 1
2 2
1 1
V1 V1
V6 V6
1 1
V4 V5 V4 V5
ANer the 3rd itera7on ANer the 4th itera7on
Distributed MST algorithms

§  each node does not need complete knowledge of the topology
§  MST is created in a distributed manner
§  e.g. GHS algorithm
Gallager, Humblet, and Spira, “Distributed Algorithm for Minimum-‐Weight
Spanning Trees,” ACM TransacRons on Programming Languages and Systems,
January 1983, pp. 66-‐67.

§  starts with one or more fragments consis7ng of single nodes
§  each fragment selects its minimum weight outgoing edge
§  using control messaging fragments coordinate to merge with a
neighbouring fragment over its minimum weight outgoing edge

§  O(|V|log|V|) 7me, provided that the edge weights are unique
§  if weights not unique, algorithm s7ll works by using the nodes IDs to break 7es
between edges with equal weight
§  requires O(|V|log|V|) + |E|) message overhead
14
Distributed MST (GHS) Algorithm (example)
V2 1 V3 V2 1 V3 Zero level
fragments
1 2 3 4 1 2 3 4
V6
V7 5 5
V1 V7
4 5 V1 4 5 V6
3 2 3 2
V4 6 V5 V4 6 V5
V2 1 1st level Nodes 3, 4, and

V3 fragments V2 1 V3
7 join 1st level
1 2 3 4 {1,2} and fragment {1,2} 1 4
2 3
5 {5,6} are
V7 V7 5
formed
V1 4 5 V6 V6
2 V1 4 5
3 3 2
V4 6 V5 V4 6 V5
Fragments {1,2,3,4,7} and V2 1 V3

{5,6} join to form 2nd level
1 2 3 4
fragment that is the MST
V7 5
V1 4 5 V6
3 2
V4 6 V5
Shortest Path Spanning Tree (SPST)

§  a shortest path spanning tree T is
a spanning tree rooted at a par7cular node, such that the
(|V|-‐1) minimum-‐weight paths from the root node to each of
the other network nodes is contained in T
2 2
2
2 6 2
6 2
3 1 3
1 5 3 1 5
2 2
2
4 4
4
Shortest Path Spanning Tree
Graph Minimum Spanning Tree rooted at vertex 1
§  note that SPST is not the same as the MST
15
Routing Applica7on
§  main func7on of the Network layer Transport
Network
§  op.mality principle:
Data Link
“if node J is on op7mal path IàK,
then op7mal path JàK is along same route” Physical
§  set of op.mal routes from all sources to a given
des.na.on form a sink tree rooted at the des.na.on
§  goal of all rou7ng algorithms is to discover and use the
sink trees for all des7na7on nodes in the network
2 24 3 2 24 3
4

1
23 9
1 4
23 9
6 18 6 18
6 6 (shortest path)
5 4 5 4
16
5
11 16 11 sink tree for
8 8 5
14
7 7 des.na.on
10 10 14
node 3
7 21 8 7 21 8
Shortest path routing algorithms

§  assume non-‐nega7ve edge weights

§  given a weighted graph (G, W) and a node s,

a shortest path tree rooted at s is a tree T such that,
for any other node v ∈ G, the path between s and v in T
is shortest path between the nodes

§  examples of algorithms which compute these shortest

path trees are Dijkstra’s and Bellman-‐Ford algorithms,
as well as algorithms which find the shortest path
between all pairs of nodes, e.g. Floyd-‐Warshall
16
Dijkstra’s Algorithm
§  each node in the network (graph) is labeled with its
distance from the source node along the best known path

§  ini7ally no paths are known: all nodes labeled with (∞,-‐)
§  as algorithm runs and paths are found, labels change to
reflect beker paths
§  once shortest possible path from source to a node is found,
its label changes from tenta4ve to permanent
Dijkstra’s Algorithm (pseudocode)

%assume s to be the root (source) node
V’ = {s}; U =V-‐{s}; E’ = φ;
for v ∈ U do
Dv = w(s,v);
Pv = s;
endFor
while U ≠ φ do
find v ∈ U such that Dv is minimal;
V’ = V’ ∪ {v}; U = U – {v};
E’ = E’ ∪ (Pv,v);
for x ∈ U do
if Dv + w(v,x) < Dx then
Dx = Dv + w(v,x);
Px = v;
endIf
endFor
endWhile
17
Dijkstra’s Algorithm (example)
1
§  assume: V2 V3
-‐ V1 is s (source node) 1 2 3 4
-‐ Dv = distance from node s to node v 5
V1 V7 V6
§  if there is no edge connec7ng two 4 4
nodes x and y, then w(x,y) = ∞ 3 2
V4 6 V5

1
§  1st mark source node V1 as permanent V2 V3
1 2 3 4
§  examine all nodes adjacent to permanent
5
node, re-‐label each with: V1 V7 V6
-‐ its distance to source node 4 4
-‐ & the permanent node from which 3 2
examined (reconstruct final path later) V4 6 V5
V’ = {1}
(1, V1) 1 (∞, V1)
V2 V3
4
1 2 3
5
V1 (∞, V1) V7 V6
4 4 (∞, V1)
3
2
V4 6 V5
(∞, V1)
(3, V1)
18
1
§  find temporary node with smallest label in V2 V3
graph and make it permanent 1 2 3 4
à becomes new working node 5
V1 V7 V6
§  examine all nodes adjacent to working 4 4
node, relabel if path to source node via 3 2
working node shorter V4 6 V5
§  REPEAT, un7l all the nodes have been

processed & their shortest distance to V’ = {1,2}
(∞, V1)
source has been computed (1, V1) 1 (2, V2)
V2 V3
4
1 2 3
(∞, V1) 5
V1 (3, V2) V7 V6
4 4 (∞, V1)
3
2
V4 6 V5
(∞, V1)
(3, V1)

1
V1 V7 V6

processed & their shortest distance to V’ = {1,2,3}
V2 V3
4
1 2 3
5
V1 (3, V2) V7 V6
4 4 (6, V3)
3
2 (∞, V1)
V4 6 V5
(∞, V1)
(3, V1)
19
1
V1 V7 V6

processed & their shortest distance to V’ = {1,2,3,4}
V2 V3
4
1 2 3
5
V1 (3, V2) V7 V6
4 4 (6, V3)
3
2
V4 6 V5 (∞, V1)
(9, V4)
(3, V1)

1
V1 V7 V6

processed & their shortest distance to V’ = {1,2,3,4,7}
V2 V3
4
1 2 3
5
V1 (3, V2) V7 V6
4 4 (6, V3)
3
2
V4 6 V5 (9, V4)
(7, V7)
(3, V1)
20
1
V1 V7 V6

processed & their shortest distance to V’ = {1,2,3,4,7,6}
V2 V3
4
1 2 3
5
V1 (3, V2) V7 V6
4 4 (6, V3)
3
2
V4 6 V5
(7, V7)
(3, V1)

1
V1 V7 V6

processed & their shortest distance to V’ = {1,2,3,4,7,6,5}
V2 V3
4
1 2 3
5
V1 (3, V2) V7 V6
4 4 (6, V3)
3
2
V4 6 V5
(7, V7)
(3, V1)
21
1
§  note that the tree computed is a shortest V2 V3
path spanning tree rooted at V1, 1 2 3 4
NOT a minimum weight spanning tree 5
V1 V7 V6
(shown below)
4 4
1 3 2
V2 V3
1 2 3 4 V4 6 V5
5
V1 V7 V6
4 4
3 2
V4 6 V5
Bellman-Ford Algorithm
§  find the shortest walk from a source node s to an arbitrary
des7na7on node v, subject to the constraints that the walk
consist of at most h hops and goes through node v only
once
22
Bellman-Ford Algorithm (pseudocode)

%assume s to be the root (source) node
Dv-‐1 = ∞ ∀ v∈ V;
Ds0 = 0 and Dv0 = ∞ ∀ v ≠ s, v ∈ V ;
h = 0;
un.l (Dvh = Dvh-‐1 ∀ v∈ V ) or (h = |V|) do
h = h + 1;
for v∈ V do
Dvh = min{Duh-‐1 + w(u,v)} ∀ u∈ V;
endFor
endUn.l
Bellman-Ford Algorithm (example)

D20=∞ D30=∞
1
V2 V3
4 D 0=∞
1 2 3 6 h = h + 1;
D10=0 5
V1 D70=∞ V7 V6 for v∈ V do
4 4 Dvh = min{Duh-‐1 + w(u,v)} ∀ u∈ V;
3 2 endFor
V5 D50=∞ endUn.l
D40=∞ V4 6
h=1 h=2 h=3 h=4

D2h
D3h
D4h
D5h
D6h
D7h
23
D2h-‐1=∞ D3h-‐1=∞
1
V2 V3
4 D
1 2 3 6 =∞
h-‐1
h = h + 1;
D1h-‐1=0 5
V1 D7h-‐1=∞ V7 V6 for v∈ V do
4 4 Dvh = min{Duh-‐1 + w(u,v)} ∀ u∈ V;
3 2 endFor
V5 D5h-‐1=∞ endUn.l
D4h-‐1=∞ V4 6
h=1 h=2 h=3 h=4

D2h=1 D3h=∞
D2h 1 1
V3
V2
D3h ∞ 1 3 4 D h=∞
2 6
D4h 3 D1h=0 5
V1 D7h=∞ V V6
7
D5h ∞ 4 4
3 2
D6h ∞
V5 D5h=∞
D7h ∞ D4h=3 V4 6

D2h-‐1=1 D3h-‐1=∞
1
V2 V3
4 D
1 2 3 6 =∞
h-‐1
h = h + 1;
D1h-‐1=0 5
V1 D7h-‐1=∞ V7 V6 for v∈ V do
4 4 Dvh = min{Duh-‐1 + w(u,v)} ∀ u∈ V;
3 2 endFor
V5 D5h-‐1=∞ endUn.l
D4h-‐1=3 V4 6
h=1 h=2 h=3 h=4

D2h=1 D3h=2
D2h 1 1 1
V3
V2
D3h ∞ 2 1 3 4 D h=∞
2 6
D4h 3 3 D1h=0 5
V1 D7h=3 V V6
7
D5h ∞ 9 4 4
3 2
D6h ∞ ∞
V5
D7h ∞ 3 D4h=3 V4 6 D5h=9
24
D2h-‐1=1 D3h-‐1=2
1
V2 V3
4 D
1 2 3 6 =∞
h-‐1
h = h + 1;
D1h-‐1=0 5
V1 D7h-‐1=3 V7 V6 for v∈ V do
4 4 Dvh = min{Duh-‐1 + w(u,v)} ∀ u∈ V;
3 2 endFor
V5 D5h-‐1=9 endUn.l
D4h-‐1=3 V4 6
h=1 h=2 h=3 h=4

D2h=1 D3h=2
D2h 1 1 1 1
V3
V2
D3h ∞ 2 2 1 3 4 D h=6
2 6
D4h 3 3 3 D1h=0 5
V1 D7h=3 V V6
7
D5h ∞ 9 7 4 4
3 2
D6h ∞ ∞ 6
V5 D5h=7
D7h ∞ 3 3 D4h=3 V4 6

D2h-‐1=1 D3h-‐1=2
1
V2 V3
4 D
1 2 3 6 =6
h-‐1
h = h + 1;
D1h-‐1=0 5
V1 D7h-‐1=3 V7 V6 for v∈ V do
4 4 Dvh = min{Duh-‐1 + w(u,v)} ∀ u∈ V;
3 2 endFor
V5 D5h-‐1=7 endUn.l
D4h-‐1=3 V4 6
h=1 h=2 h=3 h=4

D2h=1 D3h=2
D2h 1 1 1 1 1
V3
V2
D3h ∞ 2 2 2 1 3 4 D h=6
2 6
D4h 3 3 3 3 D1h=0 5
V1 D7h=3 V V6
7
D5h ∞ 9 7 7 4 4
3 2
D6h ∞ ∞ 6 6
V5
D7h ∞ 3 3 3 D4h=3 V4 6 D5h=7
25
Floyd-Warshall Algorithm
§  find the shortest path between all ordered pairs of nodes
(s,v), {s,v}∈ V
§  each itera7on yields the path with the shortest weight
between all pairs of nodes under the constraint that only
nodes {1,2,…n}, n ∈ |V|, can be used as intermediary
nodes on the computed paths
§  algorithm completes in O(|V|3) 7me
Floyd-Warshall Algorithm (pseudocode)

D = W; %W is the matrix representa7on of the edge weights
for u = 1 to |V| do
for s = 1 to |V| do
for v = 1 to |V| do
Ds,v = min{Ds,v , Ds,u+ Wu,v}
endFor
endFor
endFor
26
Floyd-Warshall Algorithm (example)
V2 D = W
2 for u = 1 to |V | do
2
2
8 for s = 1 to |V | do
6 for v = 1 to |V | do
V1 V3
1
4 Ds,v = min{Ds,v , Ds,u+ Wu,v}
endFor
4 3
1 endFor
3
V4 1
V5 endFor
5
V1 V2 V3 V4 V5
" 0 2 4 ∞ 3 % V1
$ '
$ 2 0 8 ∞ 1 ' V2
D0 = $ 6 2 0 4 3 ' V3
$ 1 ∞ ∞ 0 5 '
$ ' V4
$# ∞ ∞ ∞ 1 0 '& V5

V2 D = W
2 for u = 1 to |V | do
2
2
8 for s = 1 to |V | do
6 for v = 1 to |V | do
V1 V3
1
4 Ds,v = min{Ds,v , Ds,u+ Wu,v}
endFor
4 3
1 endFor
3
V4 1
V5 endFor
5
V1 V2 V3 V4 V5 V1 V2 V3 V4 V5
" 0 2 4 3 % V1 " % V1
$
∞
' 0 2 4 ∞ 3
$ '
$ 2 0 8 ∞ 1 ' V2 $ 2 0 8 ∞ 1 '
V2
D0 = $ 6 2 0 4 3 ' V3 D1 = $ 6 2 0 4 3 ' V3
$ 1 0 5 ' $ '
$
∞ ∞
' V4 1 3 5 0 4 V4
$ '
$# ∞ ∞ ∞ 1 0 '& V5 $# ∞ ∞ ∞ 1 0 '& V5
27
" % "
$
0 2 4 ∞ 3
' 0 2 4 8 3 %
$ '
$ 2 0 6 ∞ 1 ' $ 2 0 6 10 1 '
D2 = $ 4 2 0 4 3 ' D3 = $ 4 2 0 4 3 '
$ 1 3 5 0 4 ' $ 1 3 5 0 4 '
$ ' $ '
$# ∞ ∞ ∞ 1 0 '& $# ∞ ∞ ∞ 1 0 '
&
! ! $
0 2 4 8 3 $ #
0 2 4 4 3
&
# &
# 2 0 6 10 1 & # 2 0 6 2 1 &
D4 = # 4 2 0 4 3 & D5 = # 4 2 0 4 3 &
# 1 3 5 0 4 & # 1 3 5 0 4 &
# & # &
#" 2 4 6 1 0 & #" 2 4 6 1 0 &%
%
Distributed Asynchronous#
Shortest Path Algorithms
§  each node computes the path with the shortest weight to
every network node

§  no centralized computa.on

§  as for the distributed MST algorithm (GHS),
control messaging is required for distributed computa7on
§  asynchronous means here that there is no requirement of
inter-‐node synchroniza.on for the computa7on performed
at each node or the exchange of messages between nodes
28
Link-State vs. Distance Vector
Link-state routing
§  each node has complete network picture (topology, link costs)
§  how does each node get the global state?

§  each node reliably floods informa7on about its neighbours to every
other
§  each node independently calculates the shortest path

from itself to every other node:
§  for example by Dijkstra’s algorithm

§  in summary: each node has global knowledge of network

topology & link cost info à independently calculates routes

29
Distance vector routing
§  each node knows the links to its neighbours
§  does NOT flood this informa7on to the whole network
§  each node has provisional “shortest path”

§  e.g. node A: “I can get to node B with cost 11 via next hop node D”

§  nodes exchange this informa.on with their neighbours

§  again no flooding

§  nodes update their idea of the best path using informa7on from
their neighbours
§  in summary: nodes jointly calculate routes by itera.vely

exchanging with neighbours best-‐known routes to all nodes
Distributed Dijkstra Algorithm

§  no need to change the algorithm
§  each node periodically floods a control message throughout

the network containing link state informa.on
→ transmission overhead is O(|V|x|E|)
§  en.re topology knowledge must be maintained at each node
§  flooding of the link state informa7on allows for 7mely
dissemina7on of the topology as perceived by each node
§  each node typically has accurate informa7on to be able to
compute the shortest paths
30
Distributed Bellman-Ford Algorithm
§  assume G contains only cycles of non-‐nega7ve weight,
if (u,v) ∈ E, then so is (v,u)

§  the update equa7on is

Ds,v = minu∈N(s){w(s,u) + Du,v}, ∀v∈ V-‐{s}
§  N(s) = neighbours of s → ∀u∈ N(s), (s,u) ∈ E

§  each node only needs to know:

§  weights of the edges that are incident to it
§  the iden7ty of all the network nodes
§  es.mates (received from its neighbours) of the
distances to all network nodes
Distributed Bellman-Ford Algorithm

§  each node s transmits to its neighbours its current
distance vector Ds,V
§  likewise each neighbor node u ∈ N(s) transmits to s
its distance vector Du,V
§  node s updates Ds,v, ∀v ∈ V – {s} according to update equa7on
§  if any update changes a distance value, then s sends the current
version of Ds,v to its neighbours
§  node s updates Ds,v every .me that it receives distance vector
informa.on from any of its neighbors
§  a periodic 7mer prompts node s to recompute Ds,V or to transmit
a copy of Ds,V to each of its neighbours
31
Distributed Bellman-Ford Algorithm (example)
1 C Ini7al Ds,V
B
7 s A B C D E
A 0 7 ∞ ∞ 1
A 8 2
B 7 0 1 ∞ 8
1 C ∞ 1 0 2 ∞
E D D ∞ ∞ 2 0 2
2
E 1 8 ∞ 2 0
Ds,V
1
B C A B C D E
s
7
A 0 7 ∞ ∞ 1
A 8 2 B 7 0 1 ∞ 8
C ∞ 1 0 2 ∞
1 D ∞ ∞ 2 0 2
E D E 1 8 4 2 0
2
E receives D’s routes and updates its Ds,V

A receives B’s routes and updates its Ds,V
Ds,V
1
B C A B C D E
s
7
A 0 7 8 ∞ 1
A 8 2 B 7 0 1 ∞ 8
C ∞ 1 0 2 ∞
1 D ∞ ∞ 2 0 2
E D E 1 8 4 2 0
2
A receives E’s routes and updates its Ds,V
Ds,V
1
B C A B C D E
s
7
A 0 7 5 3 1
A 8 2 B 7 0 1 ∞ 8
C ∞ 1 0 2 ∞
1 D ∞ ∞ 2 0 2
E D E 1 8 4 2 0
2
32
A’s rou7ng table
1
B C
Des7na7on Next Hop Distance
7
B B 7
A 8 2 C E 5
D E 3
1
E E 1
E D
2
1 E’s rou7ng table
B C
7 Des7na7on Next Hop Distance
A A 1
A 8 2 B B 8
C D 4
1
D D 2
E D
2
Distance Vector Protocols

§  each node maintains a rou.ng table with entries
{des.na.on, next hop, distance (cost)}

§  nodes exchange rou.ng table informa.on with neighbors

§  whenever table changes
§  periodically

§  upon recep7on of a rou7ng table from a neighbor,

a node updates its rou.ng table if finds a “beper” route

§  entries in the rou.ng table are deleted if they are too old,
i.e. they are not “refreshed” within certain 7me interval by
the recep7on of a rou7ng table
33
Link Failure B
E
C
A
simple rerou.ng case D

§  F detects that link to G has failed F
§  F sets a distance of ∞ to G and sends update to A
§  A sets a distance of ∞ to G since it uses F to reach G G
§  A receives periodic update from C with 2-‐hop path to G (via D)
§  A sets distance to G to 3 and sends update to F
§  F decides it can reach G in 4 hops via A
Link Failure B
E
C
A
rou.ng loop case D

§  Link from A to E fails F
§  A adver7ses distance of ∞ to E
§  B and C had adver.sed a distance of 2 to E G
(prior to the link failure)
§  upon recep.on of A’s rou.ng update B decides it can reach E
in 3 hops; and adver.ses this to A
§  A decides it can read E in 4 hops; adver7ses this to C
§  C decides that it can reach E in 5 hops…
this behaviour is called count-‐to-‐infinity
34
Count-to-Infinity Problem
example: routers working in stable state
(A,1) (A,2) (A,3) (A,4)
A B C D E
(A,1) (A,2) (A,3)
rou7ng updates with distances to A shown
example: link from A to B fails
(A,2) (A,3) (A,4)
A B C D E
(A,3) (A,2) (A,3)
updated informa7on
B can no longer reach A directly,

BUT C adver7ses a distance of 2 to A
and thus B now believes it can reach A via C and adver.ses it!
35
ANer 2 exchanges of updates
(A,4) (A,3) (A,4)
A B (A,3) C (A,4) D (A,3) E

(A,4) (A,5) (A,4)
A B (A,5)
C (A,4)
D (A,5)
E

(A,6) (A,5) (A,6)
A B (A,5)
C (A,6)
D (A,5)
E
(A,6) (A,7) (A,6)
A B (A,7)
C (A,6)
D (A,7)
E

(A,8) (A,7) (A,8)
A B (A,7) C (A,8) D (A,7) E
this con.nues un.l the distance to A reaches infinity

à “bad news travels slowly” in distance vector rou7ng!
36
Split Horizon with Poisoned Reverse
§  algorithm used to avoid (not always) count-‐to-‐infinity problem
§  if A routes to C via B,
then A tells B that its distance to C is ∞
A B C
(C,∞)
B will not route to C via A if the link B to C fails
§  works for two node loops

§  does not work for loops with more than two nodes
Example where Split Horizon fails

§  when link C to D breaks,
C marks D as unreachable and reports that to A & B
§  suppose A learns it first A B
§  A now thinks best path to D is through B
§  A reports D unreachable to B and a route of cost 3 to C
§  C thinks D is reachable through A at cost 4 and reports
that to B. C
§  B reports a cost 5 to A who reports new cost to C.
§  etc...
§  core of the problem is that when X tells Y that D
it has a path somewhere, Y has no way of knowing
whether it is itself on that path!
37
Graph Theory Addendum#
“The Stable Marriage Problem”
38
The Stable Marriage Problem
§  Problem: Given N men and N Men Women

women, find a "suitable"
matching between men and
women Brek Eve
Mark Ann
§  the problem and the solu7on
can be represented as a John Lisa
bipar.te graph
Jack Helen
Bob Marge

§  graph edges represent all
possible matching
Men Women
§  matching (M) means to select
some of the edges according to
Brek Eve
some criteria
§  perfect matching: each man gets Mark Ann
exactly one woman; each

John Lisa
woman gets exactly one man
Helen
§  a matching is unstable if there is Jack
a pair who like each other more Bob Marge

than their spouses (they can
improve their situa7on by
dumping spouses and eloping)
39
Men Women
§  Gale-‐Shapley Theorem:

a stable marriage always Brek Eve
possible, found in O(n2) 7me.

Mark Ann
John Lisa
Jack Helen
Bob Marge
Matchmaker, Make Me A Match!

Gale-‐Shapley Algorithm
§  each man lists women in order of preference (from best to worst)
§  each woman lists men in order of preference
Men’s Preference List Women’s Preference List

Man 1st 2nd 3rd 4th 5th Woman 1st 2nd 3rd 4th 5th
Rick Anne Eva Ina Kate Maria Eva Vlad Rick Sam Ulrich Tim
Sam Ina Anne Eva Maria Kate Anne Tim Sam Ulrich Rick Vlad
Tim Anne Kate Maria Ina Eva Maria Sam Tim Ulrich Vlad Rick
Ulrich Eva Ina Maria Anne Kate Ina Rick Vlad Ulrich Tim Sam
Vlad Anne Ina Eva Kate Maria Kate Ulrich Sam Vesa Tim Rick
best worst best worst
40
Matchmaker, Make Me A Match!
Gale-‐Shapley Algorithm
§  all people begin unengaged
§  while there are unengaged men, each proposes un7l a woman accepts
§  unengaged women accept 1st proposal they get
§  if an engaged woman receives a proposal she likes beper, she breaks old
engagement and accepts new proposal; dumped man begins proposing
where he leN off
Men’s Preference List Women’s Preference List
Man 1st 2nd 3rd 4th 5th Woman 1st 2nd 3rd 4th 5th
Rick Anne Eva Ina Kate Maria Eva Vlad Rick Sam Ulrich Tim
Sam Ina Anne Eva Maria Kate Anne Tim Sam Ulrich Rick Vlad
Tim Anne Kate Maria Ina Eva Maria Sam Tim Ulrich Vlad Rick
Ulrich Eva Ina Maria Anne Kate Ina Rick Vlad Ulrich Tim Sam
Vlad Anne Ina Eva Kate Maria Kate Ulrich Sam Vlad Tim Rick
best worst best worst
Gale-Shapley Algorithm
Is resul7ng G-‐S matching stable?
Man 1st 2nd 3rd Woman 1st 2nd 3rd

perfect matching but unstable,
Tim Eva Anne Maria Eeva Ulrich Tim Vlad
since (Anne, Tarmo) prefer
Ulrich Anne Eva Maria Anne Tim Ulrich Vlad each other to current partners
Vlad Eva Anne Maria Maria Tim Ulrich Vlad
Tim Eeva Anne Maria Eeva Ulrich Tim Vlad perfect stable matching
(men propose)
Ulrich Anne Eeva Maria Anne Tim Ulrich Vlad
Vlad Eeva Anne Maria Maria Tim Ulrich Vlad
Reference: The Stable Marriage Problem: Structure and Algorithms (Founda7ons of Compu7ng) by
Dan Gusfield and Robert Irving, MIT Press, 1989.
41
Results of G-‐S algorithm are always stable
§  it requires two people of opposite sex in different couples to break up
a marriage
§  if a man wants to leave for another woman, then he already proposed
to her and she rejected him, so she won’t leave her husband for him

perfect matching but unstable,
Tim Eva Anne Maria Eeva Ulrich Tim Vlad
since (Anne, Tarmo) prefer
Ulrich Anne Eva Maria Anne Tim Ulrich Vlad each other to current partners
Vlad Eva Anne Maria Maria Tim Ulrich Vlad
(men propose)
Results of G-‐S algorithm are always stable
§  it requires two people of opposite sex in different couples to break up
a marriage
§  if a man wants to leave for another woman, then he already proposed
to her and she rejected him, so she won’t leave her husband for him
(women propose)
42
What happens when men do the proposing?
§  who is beker off – men or women?
What happens when men do the proposing?
§  each man has the best partner he can have in any stable marriage
§  each woman has the worst partner she can have in any stable marriage

§  note that G-‐S always produces same stable marriage
i.e. order of proposals is irrelevant
43
Moral of the story …
§  historically, men propose to women -‐ why?
§  guys: propose early and oNen
§  single ladies: ask out the guys!
END: Graph Theory Addendum#

“The Stable Marriage Problem”
44

2-Ad Hoc Networks-Graph Theory-Lecture2

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

2-Ad Hoc Networks-Graph Theory-Lecture2

Uploaded by

Copyright:

Available Formats

Ad Hoc Networks:#

Architectures & Protocols

Some Applications of Graph Theory

mechanical joints rods, beams, springs heat, energy

ﬁnancial stocks, currency transac7ons money

Application to Ad Hoc Networking

§ rou7ng protocols oNen use shortest path algorithms

§ the magnitude of graph G is characterized by:

V = { 1, 2, 3, 4}, |V| = 4

V = { 1, 2, 3, 4}, |V| = 4

Walks and paths

Walks and paths

§ a graph is called cyclic if it contains a cycle;

§ if (u, v) is an edge in a graph G,

§ a directed graph is strongly connected if A B

there is a directed path from any node

§ a graph is sparse if |E| ≈ |V|

1. G is a tree

§ for a given graph there are usually V2 V3

G = (V, E) T = (V, F) w(T) = 50

§ cluster analysis

§ approximate solu7ons to NP-­‐hard problems

§ indirect applica7ons

Prim’s Algorithm (example)

Distributed MST algorithms

V2 1 1st level Nodes 3, 4, and

Fragments {1,2,3,4,7} and V2 1 V3

Shortest Path Spanning Tree (SPST)

Shortest path routing algorithms

§ given a weighted graph (G, W) and a node s,

§ examples of algorithms which compute these shortest

Dijkstra’s Algorithm (pseudocode)

Dijkstra’s Algorithm (example)

§ REPEAT, un7l all the nodes have been

Dijkstra’s Algorithm (example)

§ REPEAT, un7l all the nodes have been

§ REPEAT, un7l all the nodes have been

Dijkstra’s Algorithm (example)

§ REPEAT, un7l all the nodes have been

§ REPEAT, un7l all the nodes have been

Dijkstra’s Algorithm (example)

§ REPEAT, un7l all the nodes have been

Bellman-Ford Algorithm (example)

h=1 h=2 h=3 h=4

h=1 h=2 h=3 h=4

Bellman-Ford Algorithm (example)

h=1 h=2 h=3 h=4

h=1 h=2 h=3 h=4

Bellman-Ford Algorithm (example)

h=1 h=2 h=3 h=4

Floyd-Warshall Algorithm (pseudocode)

Floyd-Warshall Algorithm (example)

§ how does each node get the global state?

§ each node independently calculates the shortest path

§ in summary: each node has global knowledge of network

§ each node has provisional “shortest path”

§ nodes exchange this informa.on with their neighbours

§ in summary: nodes jointly calculate routes by itera.vely

Distributed Dijkstra Algorithm

§ each node periodically ﬂoods a control message throughout

§ the update equa7on is

Ds,v = minu∈N(s){w(s,u) + Du,v}, ∀v∈ V-­‐{s}

§ N(s) = neighbours of s → ∀u∈ N(s), (s,u) ∈ E

§ each node only needs to know:

§  rou7ng protocols oNen use shortest path algorithms

§  the magnitude of graph G is characterized by:

§  a graph is called cyclic if it contains a cycle;

§  if (u, v) is an edge in a graph G,

§  a directed graph is strongly connected if A B

§  a graph is sparse if |E| ≈ |V|

1.  G is a tree

§  for a given graph there are usually V2 V3

§  cluster analysis

§  approximate solu7ons to NP-‐hard problems

§  indirect applica7ons

§  given a weighted graph (G, W) and a node s,

§  examples of algorithms which compute these shortest

§  REPEAT, un7l all the nodes have been

§  REPEAT, un7l all the nodes have been

§  REPEAT, un7l all the nodes have been

§  REPEAT, un7l all the nodes have been

§  REPEAT, un7l all the nodes have been

§  REPEAT, un7l all the nodes have been

§  how does each node get the global state?

§  each node independently calculates the shortest path

§  in summary: each node has global knowledge of network

§  each node has provisional “shortest path”

§  nodes exchange this informa.on with their neighbours

§  in summary: nodes jointly calculate routes by itera.vely

§  each node periodically ﬂoods a control message throughout

§  the update equa7on is

Ds,v = minu∈N(s){w(s,u) + Du,v}, ∀v∈ V-‐{s}

§  N(s) = neighbours of s → ∀u∈ N(s), (s,u) ∈ E

§  each node only needs to know:

§  nodes exchange rou.ng table informa.on with neighbors

§  upon recep7on of a rou7ng table from a neighbor,

this behaviour is called count-‐to-‐inﬁnity

§  works for two node loops

§  Problem: Given N men and N Men Women

§  Gale-‐Shapley Theorem: