Upload
amy-bryant
View
222
Download
0
Embed Size (px)
Citation preview
1
Architecture & Protocols for Supporting Routing &
QoS in MANET
Navid NIKAEIN
http://www.eurecom.fr/~nikaeinn
2
Outline Introduction Motivation Topology Management Route Determination Quality of Service Support Architecture Conclusion and Open Issues
3
Issues in MANET
Mobile ad hoc networks: Wireless broadcasting collision Mobility time-varying resources link Failure Lack of infrastructure fully-distributed
Complex Small devices limited resources multihop
routing
Topology Management, Routing, and QoS
4
Topology of MANET
B
A
S E
F
H
J
D
C
G
IK
Z
Y
Represents a node that has received packet P
Represents that connected nodes are within each other’s transmission range
M
N
L
5
Motivation
B
A
S E
F
H
J
D
C
G
IK
Represents a node that receives packet P forthe first time
Represents transmission of packet P
Z
YBroadcast transmission
M
N
L
6
Motivation
B
A
S E
F
H
J
D
C
G
IK
Z
Y
M
N
L
7
Motivation
B
A
S E
F
H
J
D
C
G
IK
Generate redundant and unnecessary information
Z
Y
M
N
L
8
Motivation
B
A
S E
F
H
J
D
C
G
IK
Z
Y
M
Transmissions may collide Packet P may not be delivered to node D at all, despite the use of flooding
N
L
9
Motivation
B
A
S E
F
H
J
D
C
G
IK
Z
Y
M
N
L
Flooding generates:• Redundant and unnecessary information• Collision
10
Related Work-Topology Management
Topology Management
Power Control Clustering
Non-deterministic Deterministic
Pb. Minimum dominating setNP-hard
• Core-extraction &• Spanning-tree algorithms
• DDR• TEMPOe.g. Lowest ID, Max Degree
11
Topology Management DDR- Distributed Dynamic Routing
Algorithm [2,3] Intuition Preferred neighbor election Forest Construction Zone Partitioning
Simulation Model Performance Results Summary
12
Intuition
Network Topology
Forest
TREE TREE …. TREE
ZONE ZONE …. ZONETime
BEACON
BEACON
Criteria
13
Preferred Neighbor Election
Each node selects a neighbor, called preferred neighbor, that has the maximum degree of connectivity
If identical degrees, select the one with the higher ID
a
b
e
f
c
dg
h
2
4
2
43
5
3
1
PNx={y | yNx label(y) = Max(deg(Nx) | NID) }
a
b
e
f
c
dg
h4
2
43
5
3
1
14
Forest Construction
Theorem: Connecting each node to its preferred neighbor yields to a forest [2]
Mechanism: Beaconing
a
b
e
f
c
dg
h
Forest limits the number of forwarding nodes, which reduces the redundant and unnecessary information as well as the probability of collision
a
b
e
f
c
dg
h
15
Zone PartitioningZone TreeMaintained Proactively
a
b
f
c
s
y
k
d
t
x
g
v
w
z
h
u
ei n
a
b
f
c
s
y
kd
t
x
g
v
w
z
h
u
ei n
DDR
z1
z2
z3 Zones : Improves delay performance Contributes in protocol scalability
Connected Dominating Set
16
Criteria for FC [4]:Quality of Connectivity (QoC) Buffer level b unallocated buffer
Stability level s ||
||)(
10
10
tt
tt
NN
NNxs
QoC = b + s
PNx={y | yNx label(y) = Max(QoC(Nx) | NID) }
The PNs belong to the set of high quality nodes
17
Protocol Model To study the effect of the proposed
topology management on routing performance ?
Hybrid Ad Hoc Routing Protocol (HARP) [5] Discover the shortest path Establish forward and reverse path
18
Simulation Model [Johansson, Perkins, Broach]
Traffic model: CBR 512 byte/packet 4 packets/second Source 10, 20, 30
Performance Metrics: Packet delivery fraction Avg. E2E delay Routing overhead
Movement model: Random way point [Yoon] 50 nodes 1500mx300m 0-20 m/s (or 1-20m/s) 900 simulated seconds Pause time=0, 30, 60, 150, 300,
600, 900 10 scenarios for each pause time
Beaconing: 10 s QoC = 2 s + b
19
Packet Delivery Fraction
10 sources 20 sources 30 sources
Simple routing has a slightly better PDF in low traffic load Routing+TM outperforms simple routing up to 20% as the traffic
load increases
20
Avg. E2E Delay Routing +TM significantly improves the delay performance up to 200ms
as the network conditions become stressful Shortest path is not enough
10 sources 20 sources 30 sources
21
Routing Overhead (pkt)
Simple Routing outperforms Routing+TM in low/medium traffic load Most of the packets produced by Routing+TM are the beacons
10 sources 20 sources 30 sources
22
Routing Overhead (bytes)
10 sources 20 sources 30 sources
Simple Routing outperforms Routing+TM in low/medium traffic load
23
Summary
Overall Performance
Low traffic load
Medium traffic load
High traffic load
Low MobilityFlat Routing
SimilarRouting +TM
Medium Mobility
Flat Routing
Routing +TM
Routing +TM
High Mobility Similar
Routing +TM
Routing +TM
24
Motivation for Route Selection
The effect of mobility rate and traffic load on DELAY Issue: load balancing Adjusting the weight of QoC solves the problem of load balancing
within a zone
10 sources 20 sources 30 sources
Load Balancing
25
Routing Issues
Dilemma at a node:“Do I keep track of routes to all destinations, or
instead keep track of only those that are of immediate interest?”
Three strategies: Proactive: keep track of all routes. Reactive: only those routes of immediate interest. Hybrid: partial proactive / partial reactive.
26
Related Work-RoutingAd Hoc Routing
Topology-based Position-based
Proactive Reactive Hybrid
OLSRTBRPFDSDVWRPCGSRFSH-HSRLANMAR
Proactive Reactive Hybrid
DSRAODVRDMARABRSSRTORA
ZHLSZRPCBRPHARP
DREAM LAR TerminodesGLSALM
27
Route Determination [5,6] HARP- Hybrid Ad Hoc Routing Protocol [5,6]
Intra-zone and Inter-zone routing Relative distance estimation [Aggelou,
Tafazolli] Quality of service support
Performance Results [Osafune] Summary
28
HARP: Hybrid Routing
a
b
f
c
s
y
kd
t
x
g
v
w
z
h
u
ei n
z1
z2
z3
Zone abstraction Z1
Z3
Z2
Intra-zone RoutingInter-zone Routing
mr
z4
Z4
src zone
dst zone
Shortcut intra-zone routing
- Zone level routing
29
Mechanisms to Achieve Load Balancing Inter-zone routing: route selection is done at the
destination Load balancing requires:
A set of route candidates A set of route metrics
Issues: Congestion Single route metric
Proposed mechanisms : Relative distance estimation Quality of service metrics
30
Relative Distance Estimation
SRC DST
SrcOffset
DstOffset
RD_Offset
d x R
…
Offset= mobility x elapsed time
31
Relative Distance Estimation
SRC DST
SrcOffset
Path_offset
RD_Offset
d x R
…
Offset= mobility x elapsed time
32
Query Localization Technique
dst
d=1
d=2
d=3
d=k
src
x
x
if (rd(x,dst)>rd(src,d))Drop PREQ;
else if (rd(x,dst)<TTL)TTL=rd(x,dst);
…
PREQPREQ
33
Quality of Service Support [7]
2nd Class
3rd Class
Delay
Throughput
QoSService
Best-Effort
1st Class h.(r-b)/c
QoC
c/(2h.(r-b))
1/s
Legend
h : hop count
r : buffer size
b : buffer occupancy
c : nodes’ throughput
s : stability
Application QoS Network QoC
34
Simulation Model [Johansson, Perkins, Broach]
Traffic model: CBR 512 byte/packet 4 packets/second Source 10, 20, 30
Performance Metrics: Packet delivery
fraction Avg. E2E delay Routing overhead
Movement model: Random way point [Yoon] 50 nodes 1500mx300m 0-20 m/s (or 1-20m/s) 900 simulated seconds Pause time=0, 30, 60, 150,
300, 600, 900 10 scenarios for each pause
time
35
Packet Delivery Fraction
10 sources 20 sources 30 sources
The effect of mobility and traffic load is not uniform Congestion stems from the lack of load balancing in the protocols
36
Avg. E2E Delay
10 sources 20 sources 30 sources
The effect of traffic load and mobility on the delay performance is non-uniform Load balancing :
weight of QoCRelative distance estimation
37
Avg. E2E Delay
10 sources 20 sources 30 sources
Delay performance has significant improved QoS metrics is the key factor for load balancing effect
38
Packet Delivery Fraction
10 sources 20 sources 30 sources
QoS metrics has no significant effect on PDF Route discovery and route maintenance are the key factors for improving PDF Deal with Traffic load/pattern and mobility model/rate
39
Summary
Overall Performance
Low traffic load
Medium traffic load
High traffic load
Low MobilityHARP +TM
HARP +TM
HARP +TM
Medium Mobility
HARP +TM
SimilarHARP +TM
High Mobility
HARP +TM
SimilarHARP +TM
40
Architecture
Application
HARP
DDR
Network
Topology Management
With respect to Quality of Network
Route Determination
With respect to Application requirements
AN ARCHITECTURE THAT SEPARATES [1]:
QoS Classes
QoS:Delay TPut BE
Network QualityQoC:
Power, bufferStability
41
Conclusion
Topology management improves routingLoad balancing HARP QoS metrics and RDE DDR QoC metrics
Control flooding overhead Forest redundancy & collision Relative distance estimation scope
Scalability and delay performance zone abstraction
42
ConclusionRouting requires: Adaptive topology management Load balancing Congestion avoidance mechanisms Neighboring information
Factors affecting routing performance Network size, mobility rate and model, traffic
load and pattern, network density
Traffic locality vs. network size
43
Future Work
Optimal criteria of forest Construction Introduce an adaptive routing mechanisms Effect of the factors affecting routing
performance
44
Publications[1] N. Nikaein and C. Bonnet, “An Architecture for Improving
Routing and Network Performance in Mobile Ad Hoc Network”, Kluwer/ACM MONET, 2003.
[2] N. Nikaein, H. Labiod, and C. Bonnet, “DDR-- Distributed Dunamic Routing Algorithm for Mobile Ad Hoc Networks”, MobiHoc, 2000.
[3] N. Nikaein, S. Wu, C. Bonnet and H. Labiod, “Designing Routing Protocol for Mobile Ad Hoc Networks”, DNAC, 2000.
[4] N. Nikaein and C. Bonnet, “ Improving Routing and Network Performance in MANET using Quality of Nodes”, WiOpt, 2003
45
Publications[5] Navid Nikaein, C. Bonnet and Neda Nikaein, “HARP- Hybrid
Ad Hoc Routing Protocol ”, IST, 2001.[6] Navid Nikaein, C. Bonnet, N. Akhtar and R. Tafazolli, “HARP-
v2 Hybrid Ad Hoc Routing Protocol ”, To be submitted, 2003.[7] N. Nikaein and C. Bonnet, “A Glance at Quality of Service
models in Mobile Ad Hoc Networks”,DNAC, 2002.[8] N. Nikaein, C. Bonnet, Y. Moret and I. A. Rai, “LQoS- Layered
Quality of Service Model for Routing in Mobile Ad Hoc Networks”, SCI, 2002.
[9] N. Nikaein and C. Bonnet, “ALM-- Adaptive Location Management Model Incorporating Fuzzy Logic For Mobile Ad Hoc Networks ”, Med-Hoc-Net, 2002.
46
References[Wu] DDR-Based Multicast Protocol with Dynamic
Core (DMPDC), PFHSN 2002, Eurecom Institut.[Osafune] Performance Comparison of Ad Hoc
Network Routing Protocol, IEICE 2002, Hitachi Co.[Rastogi] LQoS support for reactive ad hoc routing
protocol, Tech. Report, Indian institute of Technology.
47
Packet Delivery Fraction
10 sources 20 sources 30 sources
The effect of mobility rate and traffic load on PDF Fluctuation Congestion
CongestionFluctuation Fluctuation
48
Routing Overhead (pkt)
10 sources 20 sources 30 sources
Reaction to mobility (main cause of link failure) Caching (DSR), route request (AODV), periodical link state (OLSR), and
beaconing and PREQ (HARP+TM)
49
Routing Overhead (bytes)
10 sources 20 sources 30 sources
Reaction to mobility (main cause of link failure) Caching (DSR), route request (AODV), beaconing and PREQ (HARP+TM)
50
Forest Construction A forest is constructed by connecting each node
to its preferred neighbor and vice versa.
Node k:
PN = f then B = (ZID, k, 4, 1, f)
Node f:
PN = y then B = (ZID, f, 5, 1, y)
Periodical Beacon
INTRA-ZONE TABLE OF NODES k AND f
NID Learned_PN
f
NID Learned_PN
y
k
a
b
f
c
s
y
k
d
t
xc
g
51
Intra-Zone Clustering
Node k:
Learned_PN = c: d
B = (ZID, k, 4, 0, Learned_PN )
Learned_PN = a: b: q: y B = (ZID, k, 4, 0, Learned_PN )
Node f: Learned_PN = a: b: q: y: k
B = (ZID, f, 5, 0, Learned_PN )
Learned_PN = c: d: x: t
B = (ZID, f, 5, 0, Learned_PN )
NID Learned_PN
f a, b, q, y, t, x
c -
d -
NID Learned_PN
y x, t
k c, d
b, a, q -
INTRA-ZONE TABLE OF NODES k AND f
a
b
f
c
s
y
k
d
t
xc
g
52
Inter-Zone Clustering Either a node can succeed to add some
nodes to its intra-zone table. Otherwise, it puts the remaining nodes in
its inter-zone table.
NID NZID Z_Stability
r z4 ++
g z5 ++
INTER-ZONE TABLE OF NODE d
53
Zone Naming Select q highest ID # in intra-zone table, where
Compute a hash function on each selected ID #
separately. Concatenate all the hashed ID #.
Node k (for n=12 & d=3) :
q = 4 selected nodes: y, x, t, q
h(y)|h(x)|h(t)|h(q)
Z2= y’x’t’q’
54
Correctness Theorem: For any graph G (i.e. networks
topology), let G’ be the graph obtained by connecting each node to its PN. Then G’ is a forest.
Label(x) = Max {deg(w)|NID} wPNx
deg(xi-2) deg(xi), then xi PNxi-1 .
deg(xi-2) = deg(xi) &
NID(xi-2) NID(xi) .
55
Problem Definition ?
Trade-off between routing overhead and delay while maximizing network utilization
Trade-off between load balancing and resource conservation
Separation between topology management and route determination
What are the design elements of routing and how they must interact ?
56
Zone Behaviors
As the number of zone increases, the overhead within a zone decreases but the overhead between zones increases
Whatever the network density is, the zone diameter is bounded to 8 What is the optimal number of zones to achieve the minimum overall
overhead ? Mopt<N
Number of Zones Size of Zones
57
Zone Behaviors
Whatever the network density is, the zone diameter is bounded to 8
The average ratio of tree-path to shortest path is no longer than 2 Low Complexity O(N)
Zone Diameter Tree Path / Shortest Path
58
Stability Level Node mobility a basic characteristic of ad
hoc networks. A point of difference from wired networks. Critical to capture this node mobility in the
routing protocol in order to determine stable, reliable routes.
We propose:
||
||)(
10
10
tt
tt
NN
NNxstab
Nto: neighbors at time to.
Nt1: neighbors at time t1.
59
Stability Level
0 stab(x) 1 stab (x) = 1 high stability, stab (x) = 0 low stability
Numerator: nodes that have remained in the neighborhood of x.
High stability if none (few) of its neighbors change, low stability if many (all) of its neighbors change.
||
||)(
10
10
tt
tt
NN
NNxstab
60
Stability Example:
Stability also reflects connectivity. A large degree node, in general, more stable.
t0 t1
3/4
0
1/2
1
1
1
1
61
A path of lower hop count has higher delay, because of high buffering delays.
The other path of a possibly higher hop count, but lower delay. (also load balancing).
Example
62
Service differentiation
Outgoing traffic
Class I
Class II
Class III
60%
30%
10%
63
Status Concept A node broadcasts is QoS state itself. This QoS state reflects node’s ability to
act as a router. If selfish, then ceases to act as a router, doesn’t take part in routing protocol.
Since propagated by the node itself, the possibility of lying. (malicious behavior) pretend to be in selfish mode, to avoid being
chosen for forwarding.
64
Status Concept A selfish node does not serve the network
(does not route), hence, fair that it receives poor services from the network.
Status: the fraction of time a node has been in unselfish mode to the network.
The status of node y at node x:
][
][][
yN
yNyR u Nu [y] : number of y’s beacons in unselfish mode.
N [y] : total number of y’s beacons received.
65
Status Concept How to prevent malicious behavior? Give a status to each node, and give
incentive to a node to act as a router. If a node acts as a router, then it is not in
selfish mode. Prefer packets generated by unselfish
nodes while routing. Hence, a node that has remained selfish,
receives poor services from the network.
66
Quality of Service Definition To provide a set of service requirements
to the applications while Routing through the network, e.g. end-to-end delay.
Even Internet today, with high-speed high-quality fixed communication links, is unable to deliver guaranteed end-to-end services.
67
QoS Routing Issues
State of communication path should be considered in QoS routing: Resource availability and its stability
Cause longer path than shortest path
Trade-off between shortest path and optimal path
68
Assumption Fully symmetric environment
All nodes have identical capabilities Capabilities:
Transmission range, Battery life, processing capacity, buffer
capacity Each node periodically sends a Beacon All nodes participate in protocol operation
and packet forwarding [Michardi, Molva, Crowcroft]
69
Motivation
The low-capacity time-varying resources make maintaining accurate link state and topology information very difficult
Even Internet today, with high-speed high-quality fixed communication links, is unable to deliver guaranteed end-to-end service
70
Quality of Service Definition To provide a set of service requirements
to the applications while Routing through the network e.g. end-to-end delay
Even Internet today, with high-speed high-quality fixed communication links, is unable to deliver guaranteed end-to-end services.
71
Related Work- Quality of Service
Two examples in the wired Network: IntServ: per-flow end-to-end guarantee DiffServ: Per-class service differentiation
An example in Manet: FQMM: Flexible QoS Model for Manet
IntServ is used for high priority classes DiffServ is used for low priority classes
They require accurate link state and topology information
72
DiffServ & IntServ & FQMM
Require accurate link state and topology information
The low-capacity time-varying resources make maintaining accurate link state and topology information very difficult
Quality of service that an application requires depends strictly to the quality of network
73
Network Metrics
Hop count resource conservation
Buffer Level Stability Level
Load balancing
Trade-off between load balancing & resource conservation Compute during path discovery using concave function Reflect the quality of the communication paths Map this quality to application metrics
74
Conclusion-Architecture
Routing
Topology management Route determination
Quality of connectivity (QoC) Quality of Service (QoS)
Pro-Network Pro-Application