Towards Scale-Free Routing in MANETs

J.J. Garcia-Luna-Aceves, Stephen Dabideen, Rolando Menchcaca-

Mendez, Dhananjay Sampath, Brad Smith

University of California Santa Cruz(UCSC)

jj@soe.ucsc.edu

http://www.cse.ucsc.edu/research/ccrg/home.html

22

Proactive Routing

D

a Se

cf

h

b

Too many nodes are forced to know about how to reach each destination! Does not work well with random partitions

Path first, then data forwarding

D

Information about D

propagates away from D in a circle of

radius r

33

On-Demand Routing

D

a Se

cf

h

b

Too many nodes are forced to help find or repair ways to reach a few destinations! (RREQ flooding). Does not work with partitioned networks!

S

Too few nodes keep state for D.

So too many nodes try to fix broken paths

Information from S

propagates away from S in

a circle of radius r

Nodes with paths to D reply to S.

Path first, then data forwarding

44

Approaches

Exploit temporal and spatial locality of reference of information flows Nodes need not know about all links, nodes or clusters

in the network. Establish pre-ordering of nodes using dynamic

addresses to reduce route signaling Time and effort to establish routes is more important

than route optimality. Keep overhead increase sub-linear with number of

nodes.

Establish ordering over multiple dimensions to

provide more alternatives for routing

55

Two Approaches Today

PRIME: Protocol for Routing in Interest-defined Mesh Enclaves Exploit temporal and spatial locality of reference Nodes need not know about all links, nodes or clusters

in the network. PROSE: Positional Routing Over Searched

Elements Establish pre-ordering of nodes to reduce signaling Time and effort to establish routes is more important

than route optimality. Keep overhead increase sub-linear with number of

nodes.

66

PRIME

Nodes state their interest in certain destinations persistently.

Destinations with interest announce their presence. Only those relays between source-destination pairs

of interest incur signaling overhead. Destinations can be anything (individual nodes,

groups, content objects, roles, etc.) and any node can be a source.

Establish regions of interest (“enclaves”) for the dissemination of routing information between sources and destinations.

77

PRIME: Meshes and Enclaves

Core

MM1

R

y

w

S

p

p'

p'’

R1

z

x

MM

Boundary of the 1-extended enclave

Group receiver

Mesh member

Path node

Boundary of the unicast enclave of destination w

Boundary of the multicast enclave

Unicast source of destination w

88

PRIME Signaling

First source with interest in (unicast or multicast) sends first data packet piggybacked in a mesh-activation request (MR) MR specifies, among other fields, a horizon threshold and the

persistence of the interest Once a destination is activated with MR, it starts advertising

its existence using mesh announcements (MA). MA states: Dest ID, core ID, Dist, next hop, Seq #, and membership

Destinations, interested sources, and relays needed between them remain active for as long as there is interest in the connected component of the network.

MAs and MRs sent in HELLOs

InactiveMR

MR

MA

MA

CoreReceiverMulticast-MA with

larger id

MR

MA

Data packet

Deactivationtimeout

Deactivationtimeout

Active State

99

PRIME: Opportunistic Signaling

time

Event for group 1

Transmission of a bundle

Mesh Announcement Interval

Delay for regularevents

Events for group 1 or other groups

Event for group 3

Delay for urgentevents

Transmission of a bundle

Urgent event for group 1 or other groups or unicast destinations

No bundle is transmitted at

this time

time t’

time t’’

1010

Performance Comparison PRIME vs ODMRP+OLSR and ODMRP+AODV Assume infinite horizon and persistence for PRIME Metrics: Packet delivery ratio, Group delivery ratio, end-to-

end delay, and total overhead CBR sources at 10 pps, a packet is 256 bytes Infinite horizon and persistence! TDMA and 802.11 as MAC

Timers in ODMRP tailored to TDMA

Total Nodes 100 Node Placement Random Data Source MCBR

Simulation Time 150s MAC Protocol 802.11 Pkts. sent per src. 1000

Simulation Area 1800x1800m Channel Capacity 2000000 bps Transmission Power 15 dbm

Mobility Model Random Waypoint Pause Time 10s Min-Max Vel. 1-10m/s

Mobility Model Group Mobility Grp. Pause Time 10s Grp. Min-Max Vel. 1-10m/s

Node Pause Time 10s Node Min-Max Vel. 1-10m/s

1111

1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 60

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

Grp area 900x900m, grps of 15 nodes, 3 src per grp, 5 ucast flows

Number of concurrent active groups

Del

iver

y R

atio

PRIME mcastPRIME ucast

ODMRP with AODV

AODV with ODMRP

ODMRP with OLSROLSR with ODMRP

Delivery Ratio for Multicast and Unicast Combined (802.11 MAC)

Delivery vs. number of multicast groups:Group area 900x900, 15-node groups, 3 sources per group, and 5 unicast flows.

1212

1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 60

1000

2000

3000

4000

5000

6000Grp area 900x900m, grps of 15 nodes, 3 src per grp, 5 ucast flows

Number of concurrent active groups

Ctr

l and

Tot

al O

verh

ead

(Avg

. N

um.

of P

kts

Tx.

per

Nod

e)

PRIME: TOPRIME: CO

ODMRP and AODV: TO

ODMRP and AODV: CO

ODMRP and OLSR: TOODMRP and OLSR: CO

Overhead for Multicast and Unicast Combined (802.11 MAC)

Overhead vs. number of multicast groups:Group area 900x900, 15-node groups, 3 sources per group, and 5 unicast flows.

1313

Delivery Ratio for Multicast (TDMA MAC)

1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 60.55

0.6

0.65

0.7

0.75

0.8

0.85

0.9

0.95

1Grp area 900x900m, grps of 15 nodes, 3 sources per grp and 1 pkt each 2 sec.

Number of Concurrent Active Groups

Gro

up D

eliv

ery

Rat

io (

80%

)

PRIME

ODMRPPUMA

Group delivery vs. number of multicast groups:Group area 900x900, 15-node groups, 3 sources per group.

1414

End-to-End Delay for Multicast (TDMA MAC)

1 1.5 2 2.5 3 3.5 4 4.5 5 5.5 650

100

150

200

250

300

350

400

450

500Grp area 900x900m, grps of 15 nodes, 3 sources per grp and 1 pkt each 2 sec.

Number of Concurrent Active Groups

Ave

rage

End

-to-

End

Del

ay (

Sec

onds

)

PRIME

ODMRPPUMA

Delay vs. number of multicast groups:Group area 900x900, 15-node groups, 3 sources per group.

1515

PROSE

“Prosa” Latin for straightforward, simple Two components

Positional Labels Simple HELLO mechanism labels all nodes with positional labels

relative to one ore more elected “roots.”

Using the right DHT “Link” sources to destinations Distributed and self organizing

Routing is automatic from the positional label Overhead scalability is only O(logd N), where d = node degree, N = number of nodes

mostly due to maintaining the DHT

1616

PROSE: Positional Labels Define Routes

Root node (A) is elected in a distributed fashion using HELLOsEach node is given a label relative to node A with same HELLOsPositional labels of source and destination define the route (prefix routing)

How does node K know that node J’s label is 0210?

1717

Global ID Positional Label

Hash

DHT in PROSE

10110

D routes its mapping to its anchor’s positional label (AD)

S sends request for positional label of D to AD

Anchors store the global ID to positional label mappings.

These entries form the DHT

1818

DHT in PROSE

S learns of D’s current label and routes directly to it.

Anchor forwards packet/request to known label for D

Hashing distributes the load of anchoring

D replies to S

1919

PROSE Order Performance Signaling overhead:

Establishing labels at each node: Complexity is O(1), because each node sends HELLO to state its own label.

Publish and subscribe: Communicating ID-to-label mapping from destinations to anchors is publishing Obtaining label for destinations from anchors is subscribing Complexity is O(logd(N)), because longest path from destination to its anchor is 2

logd(N) and mappings are aggregated as they traverse the network.

Route stretch: Bounded by the amount of neighborhood routing information and the worst

prefix route Order stretch with two-hop routing information is O(logd(N+1))

Routing table complexity: Labels have length dh, with h = height of DAG Each node stores O(d2) + O(1) entries (i.e., two-hop neighbors and

destinations of interest)

2020

PROSE Performance

Qualnet Simulator 500 nodes 250 active flows Flows distributed exponentially

with mean of 1/20th the simulation duration

Simulation time = 1200s 10 Random seeds Random Waypoint mobility Pause Times varying between

1 to 10m/s Protocols compared:

AODV, OLSR, FSR

Simulation Setup900 m

600 m

2121

PROSE Performance

OLSR has heavy control overhead and tanks under high mobility

AODV suffers from constant flooding as nodes move around

FSR performs worse than AODV but better than OLSR as the scoped floods reduce interference

PROSE performs better as there is lesser interference and packets are delivered even when nodes are highly mobile.

Delivery Ratio vs. Pause Time

PROSE

2222

PROSE PerformanceControl Overhead vs. Pause Time

PROSE

2323

Degree

Lab

el

Rese

ts

PROSE Overhead: Decreases with Density

2424

PROSE Overhead: Orders of Magnitude Smaller than Traditional

Proactive and Reactive Routing

2525

Next Steps Integrate PRIME with a schedule-based MAC

Provide multicast support in PROSE

Compare PRIME and PROSE

Develop integrated PRIME and PROSE mechanisms.

Complete multi-root PROSE

Apply PRIME and PROSE mechanisms to content-based routing

QoS and multi-dimensional routing

Integrate PROSE with MAC

Provide Linux implementations of PROSE and PRIME

Make QualNet and Linux implementations available to public

2626

Publications over Past Year

Routing in Wireless Networks:

1. PRIME: Menchaca-Mendez and J.J. Garcia-Luna-Aceves, “An Interest-Driven Approach to Integrated Unicast and Multicast Routing in MANETs,” The 16th IEEE International Conference on Network Protocols (ICNP 08), Oct. 19-22, 2008, Orlando, Florida

2. D. Sampath and J.J. Garcia-Luna-Aceves. “Proactive Path Maintenance in Regions of Interest,” Proc. LOCAN 2008: 4th International Workshop on Localized Communication and Topology Protocols for Ad hoc Networks, September 29, 2008, Atlanta, Georgia.

3. BEST PAPER AWARD: R. Menchaca-Mendez and J.J. Garcia-Luna-Aceves, “Scalable Multicast Routing in MANETs Using Sender-Initiated Multicast Meshes,” Proc. IEEE MASS 2008: Fifth IEEE International Conference on Mobile Ad hoc and Sensor Systems, September 29 - October 2, 2008, Atlanta, Georgia.

4. S. Dabideen and J.J. Garcia-Luna-Aceves, “Multi-Dimensional Routing,” Proc. ANC 08: IEEE Workshop on Advanced Networking and Communications 2008, August 3–7, 2008, St. Thomas U.S. Virgin Islands.

5. B. Smith and J.J. Garcia-Luna-Aceves, “ Best-Effort Quality-of-Service,” Proc. IEEE ICCCN 2008, August 3–7, 2008, St. Thomas U.S. Virgin Islands.

6. X. Wu, H. Xu, H. Sadjadpour, and J.J. Garcia-Luna-Aceves, “Proactive or Reactive Routing: A Unified Analytical Framework in MANETs,” Proc. IEEE ICCCN 2008, August 3–7, 2008, St. Thomas U.S. Virgin Islands.

7. X. Wang and J.J. Garcia-Luna-Aceves, "Distributed Joint Channel Assignment, Routing, and Scheduling for Wireless Mesh Networks," Computer Communications, Elsevier. Accepted for publication, 2008.

8. X. Wang and J.J. Garcia-Luna-Aceves, ``Embracing Interference in Ad Hoc Networks Using Joint Routing and Scheduling with Multiple Packet Reception,'' Ad Hoc Networks, Elsevier. Accepted for publication, May 2008.

9. X. Wu, H. Sadjadpour, and J.J. Garcia-Luna-Aceves, ``A Hybrid View of Mobility in MANETs: Analytical Models and Simulation Study,'' Computer Communication, Elsevier. Invited Paper, Best Paper Series. Accepted for publication, 2008.

Thanks!