Using Directionality in Mobile Routing

1
Using Directionality in Mobile Routing

• Bow-Nan Cheng (MIT LL)
• Murat Yuksel (Univ Nevada - Reno)
• Shivkumar Kalyanaraman (IBM IRL)
• (Work done at Rensselaer Polytechnic Institute)

2
Motivation

• Infrastructure / Wireless Mesh Networks
• Characteristics Fixed, unlimited energy,
  virtually unlimited processing power
• Dynamism Link Quality
• Optimize High throughput, low latency,
  balanced load

Scalability ? Layer 3 Network Layer

• Mobile Adhoc Networks (MANET)
• Characteristics Mobile, limited energy
• Dynamism Node mobility Link Quality
• Optimize Reachability

• Sensor Networks
• Characteristics Data-Centric, extreme limited
  energy
• Dynamism Node State/Status (on/off)
• Optimize Power consumption

Main Issue Scalability
3
Scaling Networks Trends in Layer 3
Flood-based
Hierarchy/Structured
Unstructured/Flat Scalable
Mobile Ad hoc / Fixed Wireless Networks
WSR (Mobicom 07) ORRP (ICNP 06)
DSR, AODV, TORA, DSDV Partial Flood OLSR, HSLS
LGF, VRR, GPSRGLS Hierarchical Routing,
Kazaa, DHT Approaches CHORD, CAN
BubbleStorm (Sigcomm 07) LMS (PODC 05)
Peer to Peer / Overlay Networks
Gnutella
OSPF, IEGRP, RIP
OSPF Areas
Wired Networks
4
Trends Directional Communications
Directional/Directive Antennas
Hybrid FSO / RF MANETS

• Current RF-based Ad Hoc Networks
• omni-directional RF antennas
• High-power typically the most power consuming
  parts of laptops
• Low bandwidth
• Error-prone, high losses
• Free Space Optics
• High bandwidth
• Low Power
• Dense Spatial Reuse
• License-free band of operation

Omni-directional
Directional

• Directional Antennas Capacity Benefits
• Theoretical Capacity Improvements - factor of
  4p2/sqrt(ab) where a and b are the spreads of the
  sending and receiving transceiver 50x capacity
  with 8 Interfaces (Yi et al., 2005)
• Sector Antennas in Cell Base Stations Even only
  3 sectors increases capacity by 1.714 (Rappaport,
  2006)

5
ORRP Big Picture
Orthogonal Rendezvous Routing Protocol
ORRP Primitive 1 Local sense of direction leads
to ability to forward packets in opposite
directions
A
Increasing Mobility

• ORRP
• High reach (98), O(N3/2) State complexity, Low
  path stretch (1.2), high goodput, unstructured
• BUT.. What happens with mobility?

98
180o
65
S
55
T
Up to 69
42
B
2 Forwarding along Orthogonal lines has a high
chance of intersection in area
6
Mobile-ORRP (MORRP) Introduction

• What can we do?
• Replace intersection point with intersection
  region.
• Shift directions of send based on local movement
  information
• Route packets probabilistically rather than based
  on rigid next-hop paths. (No need for route
  maintenance!)
• Solution a NEW kind of routing table
  Directional Routing Table (DRT)

a
A
R
B
Introduction MORRP Key
Concepts Simulation Results
Conclusion
7
MORRP Basic Example
R Near Field DRT Region of Influence
R
S
Original Direction (a1)
S Near Field DRT Region of Influence
New Direction (a2)
D
D Near Field DRT Region of Influence

• Proactive Element Generates Rendezvous to Dest
  Paths
• Reactive Element Generates Source to Rendezvous
  Paths

Introduction MORRP Key
Concepts Simulation Results
Conclusion
8
The Directional Routing Table
Use Decaying Bloom Filter (DBF)
Routing Tables viewed from Node A
Routing Table
RT w/ Beam ID
Directional RT (DRT)
Dest ID
Next Hop
Dest ID
Next Hop
Beam ID
Dest IDs ( of Certainty)
Beam ID
C
4
B C D Z
B B Z Z
B C D Z
B B Z Z
1 1 3 3
B(90), C(30) . Z(90), D(40) .
1 2 3 4
B
1
3
A
Z
2
D
ID ID
ID set of IDs
Set of IDs set of IDs

• Soft State Traditional routing tables have a
  hard timeout for routing entries. Soft State
  decreases the level of certainty with time.
• Uncertainty with Distance Nodes closer to a
  source will have increasingly more information
  about the location of the source than nodes
  farther away
• Uncertainty with Time As time goes on, without
  updates, one will have lesser amount of
  information about the location of a node
• Uncertainty with Mobility Neighbors can
  potentially be covered by different interfaces
  based on mobility speed and direction

Introduction MORRP Key
Concepts Simulation Results
Conclusion
9
DRT Intra-node Decay
Time Decay with Mobility
Spread Decay with Mobility
a
q2 gt q1 gt q3
q2
7
q3
x
x
q1
8
a
As node moves in direction x, the certainty of
being able to reach nodes covered by region 8
should decay faster than of region 7 depending on
speed. This information is DROPPED.
As node moves in direction x, the certainty of
being able to reach nodes covered by region 2
should be SPREAD to region 1 and 3 faster than
the opposite direction. The information about a
node in region 2 should be SPREAD to regions 1
and 3.
Introduction MORRP Key
Concepts Simulation Results
Conclusion
10
MORRP Fields of Operation
S
R

• Near Field Operation
• Uses Near Field DRT to match for nodes 2-3 hops
  away
• Far Field Operation
• RREQ/RREP much like ORRP except nodes along path
  store info in Far-Field DRT

D
Introduction MORRP Key
Concepts Simulation Results
Conclusion
11
Performance Evaluation of MORRP

• Metrics Evaluated
• Reachability Percentage of nodes reachable by
  each node in network (Hypothesis high
  reachability)
• Delivery Success Percentage of packets
  successfully delivered network-wide
• Scalability The total state control packets
  flooding the network (Hypothesis higher than
  ORRP but lower than current protocols out there)
• Average Path Length
• End to End Delay (Latency)
• Aggregate Network Goodput
• Scenarios Evaluated (NS2)
• Evaluation of metrics vs. AODV (reactive), OLSR
  (proactive), GPSR with GLS (position-based), and
  ORRP under various node velocities, densities,
  topology-sizes, transmission rates.
• Evaluation of metrics vs. AODV and OLSR modified
  to support beam-switched directional antennas.

Introduction MORRP Key
Concepts Simulation Results
Conclusion
12
MORRP Aggregate Goodput Results

• Aggregate Network Goodput vs. Traditional Routing
  Protocols
• MORRP achieves from 10-14X the goodput of AODV,
  OLSR, and GPSR w/ GLS with an omni-directional
  antenna
• Gains come from the move toward directional
  antennas (more efficient medium usage)

• Aggregate Network Goodput vs. AODV and OLSR
  modified with directional antennas
• MORRP achieves about 15-20 increase in goodput
  vs. OLSR with multiple directional antennas
• Gains come from using directionality more
  efficiently

Introduction MORRP Key
Concepts Simulation Results
Conclusion
13
MORRP Simulations Summary

• MORRP achieves high reachability (93 in
  mid-sized, 1300x1300m2 and 87 in large-sized,
  2000x2000 m2 topologies) with high mobility
  (30m/s).
• With sparser and larger networks, MORRP performs
  fairly poorly (83 reach) suggesting additional
  research into proper DRT tuning is required.
• In lightly loaded networks, MORRP end-to-end
  latency is double of OLSR and about 7x smaller
  than AODV and 40x less than GPSR w/ GLS
• MORRP scales well by minimizing control packets
  sent
• MORRP yields over 10-14X the aggregate network
  throughput compared to traditional routing
  protocols with one omnidirectional interface ?
  gains from using directional interfaces
• MORRP yields over 15-20 the aggregate network
  goodput compared to traditional routing protocols
  modified with 8 directional interfaces ? gains
  from using directionality constructively

Introduction MORRP Key
Concepts Simulation Results
Conclusion
14
MORRP Key Contributions

• The Directional Routing Table
• A replacement for traditional routing tables that
  routes based on probabilistic hints
• Gives a basic building block for using
  directionality to overcome issues with high
  mobility in MANET and DTNs
• Using directionality in layer 3 to solve the
  issues caused by high mobility in MANETs
• MORRP achieves high reachability (87 - 93) in
  high mobility (30m/s)
• MORRP scales well by minimizing control packets
  sent
• MORRP shows that high reach can be achieved in
  probabilistic routing without the need to
  frequently disseminate node position information.
• MORRP yields high aggregate network goodput with
  the gains coming not only from utilizing
  directional antennas, but utilizing the concept
  of directionality itself.
• MORRP is scalable and routes successfully with
  more relaxed requirements (No need for coordinate
  space embedding)

Introduction MORRP Key
Concepts Simulation Results
Conclusion
15
Thank You!

• Questions and Comments?
• Papers / Posters / Slides / NS2 Code (MORRP,
  ORRP, OLSR AODV with Beam switched directional
  antennas)
• http//networks.ecse.rpi.edu/bownan
• bownan_at_gmail.com

Introduction MORRP Key
Concepts Simulation Results
Conclusion
16
EXTRA SLIDES
17
The Directional Routing Table
Routing Tables viewed from Node A
Routing Table
RT w/ Beam ID
Directional RT (DRT)
Dest ID
Next Hop
Dest ID
Next Hop
Beam ID
Dest IDs ( of Certainty)
Beam ID
C
4
B C D Z
B B Z Z
B C D Z
B B Z Z
1 1 3 3
B(90), C(30) . Z(90), D(40) .
1 2 3 4
B
1
3
A
Z
2
D
ID ID
ID set of IDs
Set of IDs set of IDs

• Destination ID of Certainties for each Beam ID
  stored within a Decaying Bloom Filter
• Bloom Filter A space-efficient probabilistic
  data structure that is used to test whether an
  element is a member of a set.
• Consist of a bit array and a set of k linearly
  independent hash functions
• Storage IDs are hashed to each of the k hash
  functions ? stores a 1 in position in the bit
  array for each hash function.
• Search IDs are hashed through each of the k hash
  functions ? if all positions have a 1, then the
  ID is in the set. Otherwise, the ID is not in the
  set

18
DRT Decaying Bloom Filter Primer
ID 1
ID 2
ID 6
4 Hash Funcs
h1(x) (x2 20) 32
h2(x) x 32
h3(x) (x 5) 32
h4(x) (x3 25) 32
h2(1) 1
h3(1) 6
h4(1) 26
h1(2) 24
h2(2) 2
h3(2) 7
h4(2) 1
h1(6) 24
h2(6) 6
h3(6) 11
h4(6) 17
h1(1) 21
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32 Bit Array
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
1
1
1
1
1
1
1
DRT
What policies For decaying bits can we employ?
Traditional Bloom Filter
Search ID 1 4 of 4 bits match (IN set)
Search ID 6 2 of 4 bit match (Not in set)
Decaying Bloom Filter (DBF)
Search ID 1 4 of 4 bits match (100 chance in
set)
Search ID 6 2 of 4 bit match (50 chance in set)
19
DRT Inter-Node Decay
A
S
Strong Info
C
A
B
Decay 50 of Bits
B is now 100 sure A is 1 hop away while only 50
sure C can be reached through sending out
interface 1
20
DRT Intra-node Decay
Time Decay with Mobility
Spread Decay with Mobility
a
q2 gt q1 gt q3
q2
7
q3
x
x
q1
8
a
As node moves in direction x, bits in DBF of
region 8 should decay faster than of region 7
depending on speed
As node moves in direction x, bits in DBF of
region 2 should be SPREAD to region 1 and 3
faster than the opposite direction
Beam ID 1
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
Beam ID 2
0
1
0
0
0
0
1
0
0
0
0
0
0
1
0
0
0
0
1
0
0
0
0
Beam ID 3
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
0
21
Conclusion / Future Work

• Used Directionality to scale wireless networks
  (ORRP, MORRP)
• Used concept of Virtual Directions to scale
  overlay networks (VDR)
• Future Work Extensions
• Virtual direction abstraction analysis
• Hybrid ORRP (that works with omnidirectional and
  directional antennas)
• Analysis of Effect of knobs in MORRP
• New Directions with Directionality
• Multi-path / multi-interface diversity
• Directional Network Coding
• Destination-based routing based on local
  directions

22
Scaling Networks OSI Model
Transport Layer Handles reliable transmissions
end-to-end
Network Layer Manages routing from end-to-end
Layers 5-7
4 Transport Layer
3 Network Layer
2 Link Layer
Link Layer Manages node-to-node transmissions
1 Physical Layer
Physical Layer Handles transmission of bits
through a medium
Application/Presentation/Session Layers Deal
with the actual programs/data
23
Research Objectives

• Wireless Mesh Context
• Can directionality be used to address issues with
  scalability at higher throughput in layer 3
  routing?
• Mobile Ad Hoc Context
• Can directionality be used to address issues with
  high mobility and throughput in layer 3 routing?
• Overlay Network Context
• Can directionality be used to scale flat,
  unstructured networks?

24
Orthogonal Rendezvous Routing Protocol
?
(4,6)
D
By removing position information, can we still
efficiently route packets?
(8,5)
(15,5)
S
D(X,Y)?
(0,4)
(12,3)
(5,1)
Issues in Position-based Schemes
25
ORRP Design Considerations

• Considerations
• High probability of connectivity without position
  information Reachability
• Scalability O(N3/2) total state information
  maintained. (O(N1/2) per node state information)
• Even distribution of state information leading to
  no single point of failure State Complexity
• Handles voids and sparse networks
• Trade-offs
• Path Stretch
• Probabilistic Reachability

26
ORRP Proactive and Reactive Elements
North
North
North
120o
North
A to D
North

• ORRP Announcements (Proactive) Generates
  Rendezvous-to-Destination Routes
• ORRP Route Request (RREQ) Packets (Reactive)
  Generates Source-to-Rendezvous Rts
• ORRP Route Reply (RREP) Packets (Reactive)
• Data path after route generation

27
Reachability Numerical Analysis
Punreachable Pintersections not in
rectangle
Probability of Unreach highest at perimeters and
corners
NS2 Simulations with MAM show around 92
reachability
4 Possible Intersection Points
57
98.3
99.75
67.7
28
Path Stretch Analysis

• Average Stretch for various topologies
• Square Topology 1.255
• Circular Topology 1.15
• 25 X 4 Rectangular 3.24
• Expected Stretch 1.125

x 1.255
x 1.15
x 3.24
29
State Complexity Analysis/Simulations
ORRP states are distributed fairly evenly in an
unstructured manner (no single point of failure)
ORRP state scales with Order N3/2
30
ORRP Simulation Results Summary

• Base Case
• Reach 99 for Square topologies, 92 for
  Rectangular topologies (MAM helped)
• Path Stretch Roughly 1.2
• Goodput About 30x more aggregate network
  goodput than AODV, 10x more aggregate network
  goodput than OLSR and 35x more aggregate network
  goodput than GPSR with GLS (due to better usage
  of medium)
• Network Voids
• Average path length fairly constant (Reach and
  State not different)
• Additional Lines
• Reach/Path Stretch All showed large gains from
  1 to 2 lines but diminishing returns thereafter
• Goodput Higher average network throughput with
  additional lines (better paths and higher reach)
  but not by much
• Varying Number of Interfaces
• Significant increase in reachability from 4 to 8
  interfaces, but gains trail off

31
ORRP Summary

• ORRP achieves high reachability in random
  topologies
• ORRP achieves O(N3/2) state maintenance
  scalable even with flat, unstructured routing
• ORRP achieves low path stretch (Tradeoff for
  connectivity under relaxed information is very
  small!)
• ORRP achieves roughly 30X in aggregate network
  goodput compared to AODV, 10X the aggregate
  network goodput compared to OLSR, and 35X the
  aggregate network goodput compared to GPSR with
  GLS.
• Relevant Papers
• B. Cheng, M. Yuksel, and S. Kalyanaraman,
  Rendezvous-based Directional Routing A
  Performance Analysis, In Proceedings of IEEE
  International Conference on Broadband
  Communications, Networks, and Systems
  (BROADNETS), Raleigh, NC, September 2007.
  (invited paper)
• B. Cheng, M. Yuksel, and S. Kalyanaraman,
  Directional Routing for Wireless Mesh Networks A
  Performance Evaluation, Proceedings of IEEE
  Workshop on Local and Metropolitan Area Networks
  (LANMAN), Princeton, NJ, June 2007.
• B. Cheng, M. Yuksel, and S. Kalyanaraman,
  Orthogonal Rendezvous Routing Protocol for
  Wireless Mesh Networks, Proceedings of IEEE
  International Conference on Network Protocols
  (ICNP), pages 106-115, Santa Barbara, Nov 2006.

32
Wireless Nets Key Concepts to Abstract

• Directionality CAN be used to provide high reach,
  high goodput, low latency routing in wireless
  mesh (ORRP) and highly mobile adhoc networks
  (MORRP)
• Primitives
• Local directionality is enough to maintain
  forwarding along a straight line
• Two sets of orthogonal lines intersect with a
  high probability in a bounded region
• Overlay Networks
• Can we take these concepts to scale unstructured,
  flat, overlay networks?

33
Virtual Direction Routing Introduction

• Structured vs. Unstructured Overlay Networks
• Unstructured P2P systems make little or no
  requirement on how overlay topologies are
  established and are easy to build and robust to
  churn
• Typical Search Technique (Unstructured Networks)
• Flooding / Normalized Flooding
• High Reach
• Low path stretch
• Not scalable
• Random Walk
• Need high TTL for high reach
• Long paths
• Scalable, but hard to find rare objects
• Virtual Direction Routing
• Globally consistent sense of direction (west is
  always west) ? Scalable interface to neighbor
  mapping
• Routing can be done similarly to ORRP
• Focus (for now)
• Small world approximations

Virtual Direction Routing
Random Walk
34
VDR Neighbor to Virtual Interface Map
Example Neighbor IDs used Instead Of SHA-1
Hashes
30 8 6
26
15 8 7
30
15
10
10 8 2
10
1
15
68
26
30
26 8 2
68
68 8 4

• Neighbors are either physical neighbors connected
  by interfaces or neighbors under a certain RTT
  latency away (logical neighbors)
• Neighbor to Virtual Interface Mapping
• Each neighbor ID is hashed to 160 bit IDs using
  SHA-1 (to standardize small or large IDs)
• The virtual interface assigned to the neighbor is
  a function of its hashed ID (Hashed ID number
  of virtual interfaces)

35
VDR State Seeding and Route Request
State Seeding State info forwarded in
orthogonal directions, biasing packets toward IDs
that are closer to SOURCE ID. Packets are
forwarded in virtual straight lines.
10 1 9 26 1 25
10
14 1 13 22 1 21
5 1 4 13 1 12
5
14
Ex Seed Source Node 1
10
Route Request RREQ packets are forwarded in
orthogonal directions, biasing packets towards
REQUESTED ID
10 12 2 26 12 15
6
5 12 7 13 12 1
6 12 6 38 12 26
Ex Route Request Node 12 RREQ Source Node 1
13
36
VDR Simulation Parameters
46
Flooding
68
5
RREQ Path
Normalized Flooding
Rendezvous Node
10
6
30
13
1
26
38
RREP Path
2
Virtual Direction Routing
RREQ Node 12
48
Seed Path
VDR Route Request
12
Virtual View
VDR Random NB Send (VDR-R)

• Simulation of VDR vs. RWR, VDR-R
• VDR-R VDR with random neighbor forwarding (no
  biasing)
• RWR Data is seeded in 4 random walks and 4
  walkers are sent for search
• PeerSim 50,000 Nodes, Static Dynamic Network
• Reach Probability High (98 w/ TTL of 100)
• Average Path Stretch High (16)
• State and Load Spread Not evenly distributed

Random Walk Routing (RWR)
Random Walk
37
VDR Robustness Results

• State Distribution Network-wide
• Average States maintained relatively equal for
  VDR, VDR-R and RWR at 350-390
• VDR States are not very evenly distributed, with
  some nodes having more state than others. This is
  due to the sending bias

• Robustness to Network Churn
• VDR drops only 5 compared to VDR-R and RWR which
  drop 12-15 reach when going from 0 to 50
  network churn
• Even with a TTL of 50, VDR reaches a good amount
  of the network

5 drop
15 drop
12 drop
38
VDR Key Contributions

• Introduction of the concept of Virtual Directions
  to eliminate need for structure (coordinate
  space, DHT structures) to scale flat,
  unstructured overlay networks
• A flat, highly scalable, and resilient to churn
  routing algorithm for overlay networks
• VDR provides high reach (98 even only for a TTL
  of 100 in a 50,000 node network)
• VDR drops only 2-5 going from 0 churn to 50
  churn
• Relevant Papers
• B. Cheng, M. Yuksel, and S. Kalyanaraman, Virtual
  Direction Routing for Overlay Networks, In
  preparation for submission to IEEE Peer to Peer
  Computing (P2P) 2008.