Collaborative Similarity Measure for Intra-Graph Clustering
Final
1. Dissertation Defense
On-Demand Link-State Routing in
Ad-Hoc Networks
Soumya Roy
Computer Engineering,
University Of California, Santa Cruz
Adviser : Prof. JJ Garcia Luna Aceves
4. Wireless Ad Hoc Networks
WLAN Ad hoc Networks
Nodes acts as sources, relays and destinations of data packets
Routing protocol is needed for data delivery
5. Challenges in Designing Routing
Protocols for Mobile Ad-Hoc Networks
Significant Packet Loss :
Fading, interference, collisions
Less bandwidth than wired networks
Network Topology highly dynamic
because of router or host mobility
Routing protocols should adapt fast to
link failures and converge fast
Routing overhead should be minimal
6. Taxonomy of Routing Protocols
Proactive Routing (WRP, DSDV, OLSR,
STAR)
Maintains routes for all destinations
Redundant routes
Reactive Routing (AODV, DSR, DST)
Maintains routes for necessary nodes
Uses flooding of requests to discover paths
Long delays for path set up
7. Main goal is to ….
Explore how link-state information can be
used efficiently for setting up
on-demand routes
8. Previous Work
Proactive link-state protocols
Source Tree Adaptive Routing (STAR)
Exchanges source trees
Topology Broadcast based on Reverse-Path
Forwarding (TBRPF)
Exchanges reportable part of source tree
Optimized Link-State Routing (OLSR)
Uses multipoint relays to prevent flooding
Performance good compared to on-demand
routing protocols using routing information in
other forms
12. What is Source Tree
Each node can build source tree based
on paths to all reachable destinations
a
b c
SOAR exchanges minimal source trees
d e f
g h i j k
l m
Source Tree at node a
13. What is a Minimal Source Tree
Contains paths to important nodes
(relays, potential relays, receivers)
a a
b b c
c
d e f d e f
jj g jj
g h i k
Minimal Source Tree
l m l
reported by node a
Source Tree at a
14. Construction of Topology Table
i
x a a x
i y c i c y
+ =
Minimal source tree Minimal source tree Partial topology at i
of x of a
Modified Bellman-Ford Algorithm used
for path selection
15. Summary
Query, Reply for path discovery
Updates for path repair
ForcedUpdate, ForcedReply for forcing
nodes adopt shortest paths
Links validated using sequence number
Data packets contain traversed paths
16. Performance Evaluation (ns2)
20 nodes
1000mx300m rectangular field
IEEE 802.11 MAC
Link layer notification of link failures
Re-routing of data packets possible in
case of link breakage
Comparisons with DSR and AODV
21. Path Selection Algorithm
Links advertised by b
(c) (c)
c d f Links advertised by c
(c)
(c) (c)
a (b) (c)
g
(b) h
(b) e
b
Path through a neighbor valid if later has
advertised it
Choose the shortest among the valid options
22. Computing Source Tree
OPT-TREE : Given a graph and a set of
labels corresponding to each link, what
is maximum size of the source tree that
can be built while satisfying the rules of
shortest path and on-demand route
advertisement?
NP complete problem (reduction of
3SAT problem to a graph)
24. Heuristic for Computing Source Tree
A polynomial time approximation (O(nd^2))
algorithm
Finding shortest valid paths for a destination
and combining those paths
Local maxima instead of global maxima
Disadvantage: computing source tree not an
efficient usage of routing information
A mechanism for policy-based routing
25. Computing source graph (in SOAR)
MBF BF
(a,d1) (x,d2)
j d1 >d2
(x,d2)
i
i i
k (a,d1+1)(x,inf) (x, d2+1)
a x a x a x
d1 d2
j (j,k) link has been j j
advertised by x only
k k k
Partial topology at i
Source Graph at i Advertised
Source Tree at i
26. How does SOAR handle path
selection problem
Computes source graph
Exchanges source trees
Uses forced routing to convert source
graphs to trees
Disadvantage:
Loops can form
Extra overhead of forced routing
27. Next Step …….
Use of complete path information
Links in paths validated using sequence
numbers
Links merged to help in local route
repair and limited route requests
Instantaneous loops to be prevented
without any control overhead in data
packets or any global synchronization
32. RREQ
d
a
RR
EQ
RREQ
i
j
k
RR
EQ c
b
RREQ
RREPACK
RREP
a
d
RR
EP
RREP
Example (Route Discovery)
i
j
k
RR
RREPACK PE
RR AC
b
K
c
EP
RREP
RREPACK
33. Example (Route Error and Repair)
j
j
d c
d c
RREQ
FRREQ
RREQ
FRREP
a b
a b
FR
EQ
RR
FR
RE
RERRACK
RR
EQ
RE
P
k
RERR
k
Q
i RREQ i
34. Properties of OLIVE
When cost of path can increase, inter-
neighbor synchronization is used
Predecessors release the successor
Loop-free paths set up when all
predecessors release
Every node knows correctly at every
instance its set of predecessors
Data packets from unknown
predecessors always dropped
35. Performance Evaluation (ns2)
50 nodes
1500x300m (range = 250m)
2Mbps DSSS radio interface
Random Waypoint model
IEEE 802.11
Compared with DSR, OLSR, TBRPF,
AODV
41. Summary
Elaborately studied the problem of
using on-demand link-state information
for routing
Developed routing solutions that give
better performance than existing
popular routing protocols
OLIVE can be very good choice for use
in ad hoc networks
43. Outline
Introduction
Reasons behind node-centric hybrid
routing
Methods and new protocols
Performance evaluation
44. Introduction
Practical scenarios for mobile ad-hoc
networks will have traffic between
mobile nodes and netmarks
Where can we see such scenarios :
Internet Access Points
Hosts for DNS services
Web Proxies
Group leaders
Cluster-heads for hierarchical routing
46. Related Work in Hybrid Routing
Zone Routing Protocol (Haas & Pearlman)
Proactive routing within zones
Reactive routing between zones
Landmark Hierarchy (P.F. Tshuchiya)
Node centric approach to hierarchical
routing for proactive routing
Landmark’s address: address for common
node
Our approach: Type of routing is based
on nodes and not on zones or areas
47. Why Node-centric Hybrid Routing
On Demand path creation towards netmarks
Each session set up leads to flooding, delay
Intuitively not best approach if the netmark is
communicated frequently though not continuously
Maintaining pro-active routes
High control overhead
Redundant paths as peer-to-peer communication is
much less compared to number of nodes
Node-centric hybrid routing: a tradeoff
Proactive routes towards netmarks
On demand path creation among mobile nodes
48. Methods of Node-Centric Hybrid
Routing
Prolonged Caching
Caching of netmark information for long
enough periods
Proactive routes
Proactive routing for the netmark so that
paths towards netmarks remain always
updated while paths between mobile nodes
are set up on-demand
49. Protocols with Prolonged Caching
Route errors and route requests for netmarks
do not depend on traffic
AODV : longer routing entry timeout for
netmarks
DSR : advantage not realizable since routing
solely controlled by traffic
SOAR :Netmarks will be important for longer
time interval, hence caches/maintains
netmark path information for longer periods
50. Protocols With Proactive Netmark Routes
1. Nemarks advertise their presence
MAC layer sends beacons
Routing Layer sends Hellos
2. Route errors and route requests for
netmarks do not depend on traffic
Changes 1 and 2 can be easily made
for DSR, SOAR and AODV
51. Protocols with Proactive Routes
(contd…)
3. New paths to netmarks are always
advertised
Change 3 needs modifications
SOAR sends updates immediately on
discovering new routes
AODV and DSR need new type of control
packets for proactive route set up
52. SOAR with Hybrid Functionality
SOAR performs better than both DSR or AODV
Modifications of SOAR to incorporate hybrid
routing simpler than in DSR or AODV
Netmark Aware On Demand Link-State Routing
(NOLR) : extended caching of netmarks
Netmark-Enhanced Source Tree Routing
(NEST) : proactive routes for netmarks
53. Setting up paths in NEST
Forward paths are set up proactively
Reverse paths are set up by traffic flows (soft state or
on-demand) (e,b)
hel (e,a)
lo n
n
lo
n
hel
b b b
(e,d)
update
a a (e,c) a
(e,c)
update d
upd
(e,e) d d
(e,e)
at e
c c c
update
e e e
54. Issues in Hybrid Routing
Forwarding of data packets
Use of subnet address for ad hoc networks
Multiple Netmarks
Load balancing
Complexity in packet forwarding (static, dynamic,
hybrid affiliations)
Asymmetry in paths
Change in query mechanism (anycast queries)
56. Traffic and Mobility Models
Mobility Model : random waypoint model
Traffic Pattern
FLOW OFF/ON model
Data Traffic and Voice traffic Simulation
57. Simulation Scenario 1
31 nodes
single netmark in the centre of a
rectangular field
(1000m x 500m) size
Range = 250m
Simulation length : 600 secs
load = 3 packets/sec 5 packets/sec
Speed : 5m/s-20m/s
60. Simulation Scenario 2
31 nodes
single netmark
Static
Mobile in a restricted region
Mobile throughout the rectangular field
Traffic flows ( 2 models)
6 random flows among netmark and mobile nodes
(battlefield or relief)
Any mobile node can talk to netmark (Internet)
64. Contributions
First detailed study on use of link-state
information on-demand
Developed two unique highly efficient
routing protocols
Demonstrated the complexity of path
selection in on-demand routing
Presented a new genre of hybrid
routing targeting practical ad-hoc
networks
65. Future Work
Opportunistic version of OLIVE to target
practical ad-hoc networks
Node-centric version targeting huge
networks without added complexity of
hierarchy management
Modeling ad-hoc networks based on
parameters like node density
66. Acknowledgements
Thanks to JJ for being the greatest adviser
Thanks to Richard, Katia for being on my
defense committee, and Suresh for being on
my advancement committee.
Thanks to my parents and brother who have
always supported me
Thanks to my friends in Baskin Engineering,
Bay Area and CCRG for all the help during
my PhD life