Download Broadcasting in Ad Hoc Network

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project

Document related concepts

Wake-on-LAN wikipedia , lookup

IEEE 1355 wikipedia , lookup

CAN bus wikipedia , lookup

Power over Ethernet wikipedia , lookup

Transcript
Distance ADaptive (DAD) Broadcasting
for Ad Hoc Networks
The Paper

X. Chen, M. Faloutsos & S.V.
Krishnamurthy, “Distance Adaptive
Broadcasting for Ad Hoc Networks”,
in IEEE MILCOM 2002.
Copyright: S.
Krishnamurthy, UCR
Objective

To find a good way to perform
effective broadcasting in an ad hoc
network such that:
Fewer number of rebroadcasts are
needed.
 Achieve a higher coverage
 Achieve power efficiency
 Result in a fewer number of collisions

Copyright: S.
Krishnamurthy, UCR
Our Approach

Only outmost nodes rebroadcast
Outmost nodes are more likely to reach
new nodes
 We achieve a reduction in contention
 Each node identifies the outmost nodes
within its range based on some
neighborhood information that is
exchanged.

Copyright: S.
Krishnamurthy, UCR
Roadmap
Description of problem
 Metrics of Interest
 Previous work
 Our approach
 Results and Discussion
 Future work

Copyright: S.
Krishnamurthy, UCR
Broadcasting in Ad hoc network

Definition



A session in which information is to reach all
nodes.
Multiple rebroadcasts (local) would be needed
Objective

Perform broadcasting in an efficient way so as
to use fewer rebroadcasts while maintaining
the requisite coverage
Copyright: S.
Krishnamurthy, UCR
Metrics and parameters

Metrics
Coverage – fraction of nodes reached
 Broadcast Efficiency
 Broadcast Latency or Duration


Parameters
Mobility -- speed
 Node density

Copyright: S.
Krishnamurthy, UCR
Previous work
Flooding
 Expanding Ring Search –
application specific.
 S.Y.Ni, Broadcasting storm problem


Probabilistic Scheme wherein a node
re-broadcasts a received packet with a
certain probability.
Copyright: S.
Krishnamurthy, UCR
General probabilistic broadcast (GEN)
Parameter k as the target rebroadcast
size
 When node receives a packet, it
randomly generates a number n that
is between 0 and its neighborhood size
 If n < k, it will rebroadcast, otherwise
it discards the packet.

The protocol attempts to have ‘k’ new
rebroadcasts for every broadcast.
Copyright: S.
Krishnamurthy, UCR
Broadcasting and outmost nodes




Not every node is needed
to rebroadcast
It’s more efficient to let the
outmost nodes rebroadcast
Outmost nodes span the
desired area more quickly
E.g. outmost nodes
4,5,6,7,8 rebroadcast, it is
not necessary for nodes
1,2,3 to rebroadcast.
Copyright: S.
Krishnamurthy, UCR
Our approach
Using power level to decide outmost
nodes
 Distance ADaptive: based on local
information, a node decides certain
number of outmost nodes that are to
rebroadcast.
 Two variants  DAD-NUM, DAD-PER

Copyright: S.
Krishnamurthy, UCR
DAD-NUM
• Fixed number of outmost nodes performing
rebroadcast
• Node keeps a neighbor table to records the received
power level from each neighbor.
• This table is sorted to decide the threshold power
level that identifies the outmost nodes
• Include this threshold in the broadcast packet
• When the packet is received, the receiver compares
the threshold in the packet and the received power
strength to decide whether it should rebroadcast
Copyright: S.
Krishnamurthy, UCR
DAD-NUM State diagram
Init_state
Finished
Received
a packet
Pkt_recv.
Set timer
Find threshold power and put
it in packet, broadcast packet.
Time out
Time out
Pkt_Gen
If receiving power is less than
threshold power in packet, ignore packet.
Copyright: S.
Krishnamurthy, UCR
DAD-PER

Only difference from DAD-NUM:
• A fixed percentage of total neighboring
nodes performing rebroadcast

Not good for topologies wherein node
density is small or variant.
Copyright: S.
Krishnamurthy, UCR
Simulation results

System Setup
• Glomosim 2.0
• 802.11 MAC CSMA/CA
• Hello Message: every 5 seconds
• Network topology in 3000m x 3000m
• Transmission radius 223m
• Result is average on 200 random topologies
Copyright: S.
Krishnamurthy, UCR
Broadcast efficiency


DAD-NUM has the
highest efficiency
DAD-PER is better
than GEN when the
rebroadcast size is
small.
Copyright: S.
Krishnamurthy, UCR
Coverage


Bars represent the
improvement in
coverage of DADNUM over GEN.
DAD-NUM can
achieve up to a
20% increase in
coverage.
Copyright: S.
Krishnamurthy, UCR
Efficiency v.s. Coverage

DAD-NUM can
achieve a better
Coverage than GEN
while attaining a
higher broadcast
efficiency.
Copyright: S.
Krishnamurthy, UCR
Latency


DAD-NUM takes a
short time to
complete the
broadcast session
than GEN
Improvement can
be up to 21%
Copyright: S.
Krishnamurthy, UCR
Conclusion and future work

Conclusion

DAD is better than GEN with regards
to:
• Coverage
• Efficiency
• Latency

Future work

Apply DAD to in power-heterogeneous
ad hoc networks.
Copyright: S.
Krishnamurthy, UCR
Distributed Power Control in Ad Hoc
Networks
Copyright: S.
Krishnamurthy, UCR
The Paper

S.Agarwal, S.V.Krishnamurthy,
R.H.Katz and S.Dao, “Distributed
Power Control in Ad-hoc Wireless
Networks”, IEEE PIMRC 2001.
Copyright: S.
Krishnamurthy, UCR
The IEEE 802.11 MAC
•RTS – CTS – DATA – ACK
•Solves the hidden and exposed terminal problem
in most cases.
E
C
RTS
A
B
CTS
D
Copyright: S.
Krishnamurthy, UCR
Why is Power Control Hard?
No centralized controller as in
cellular networks.
 Distributed decisions on what power
to use.

Copyright: S.
Krishnamurthy, UCR
Benefits
Energy Conservation
 Frequency Re-use – more number of
simultaneous transmissions possible
– translates into an increase in the
network capacity.

Copyright: S.
Krishnamurthy, UCR
Transmission Range Models
Typically models assume a circular
range – 250 meters is the
transmission range – within this
range, data can be decoded.
 Interference range – larger than the
transmission range – data cannot be
decoded – only the interference can
be sensed.

Copyright: S.
Krishnamurthy, UCR
Clustering
Elect clusterheads for a group of
nodes.
 The clusterhead is responsible for
the transmit power for each node
within its cluster.
 Imposing a cellular infrastructure
onto an ad hoc framework.
 Refer to paper for reference.

Copyright: S.
Krishnamurthy, UCR
Power Control Extensions to
the IEEE 802.11 MAC
Ten Quantized Power Levels
 The levels vary linearly  the
difference between levels is about a
tenth of the maximum power level.
 Implement a power control loop
between a communicating pair.

Copyright: S.
Krishnamurthy, UCR
Modifications to control
messages



RTS/CTS messages modified to include a
new field.
When a node receives the RTS message it
measures received signal strength (There
is usually what is called a Received Signal
Strength Indicator or RSSI in hardware).
The receiver indicates the ratio of the
received strength to the minimum
acceptable strength in the CTS header.
Copyright: S.
Krishnamurthy, UCR
The Power Loop Closed



The transmitter (the originator of data)
then does a similar computation with the
received CTS message.
It then includes a similar ratio in the
header of the DATA message.
So both the transmitter and the receiver
are now aware of the power situation on
the link – how well are we doing!
Copyright: S.
Krishnamurthy, UCR
Basic Idea



Increase power if the power requirements
are not satisfied – packet loss.
Decrease power if power requirements
are satisfied 
Maintain table for each neighbor – to
know the power level to be used in order
to communicate with that neighbor.
Copyright: S.
Krishnamurthy, UCR
Nuances





A single power measurement will not
suffice.
One would need to dampen fluctuations.
Once a power level is chosen, ten
transmissions at that level (a heuristic
parameter).
The power control loop is only used for
unicast transmissions – routing updates
etc. that are broadcast do not use this.
For further details – read paper.
Copyright: S.
Krishnamurthy, UCR
Sample Simulation Results
Simulations were done in ns 2.0
 Various mobility models were
considered.
 TCP Throughput (actually goodput –
does not take into account
duplicates) is the metric of interest.

Copyright: S.
Krishnamurthy, UCR
Parameters
Copyright: S.
Krishnamurthy, UCR
Sample Simulation Results
• Througput
improvement is due
to an increase in
capacity – higher
frequency re-use.
Copyright: S.
Krishnamurthy, UCR
Sample Simulation Results
(Cont).
• Decrease in overall energy consumption
(on average).
Copyright: S.
Krishnamurthy, UCR
Why can performance be
worse ?
Data
RTS
1
3
2
4
•Node 1
establishes a high
power link
•Node 3 is
receiving
Data from
Node 4
CTS
Node 2 does not know about the data transfer
The high power CTS collides with the Data at Node 3
Copyright: S.
Krishnamurthy, UCR
Power Control leads to
Asymmetry
There is an inherent asymmetry
resulting from power control.
 Simply changing power levels can
lead to unfairness – and collisions
and can in some scenarios degrade
performance.

Copyright: S.
Krishnamurthy, UCR
Problems at the routing
layer
Traditional routing protocols may no
longer be used.
 Uni-directional links are formed.
How can they be used ?
 Neighbor discovery a challenge.

Copyright: S.
Krishnamurthy, UCR
Interesting topics for
projects






Few papers try to do power control
Paper by Monks in INFOCOM 2001
Paper by Jung and Vaidya – Mobicom
2002.
However, no capacity increase, use
highest power for transmission of control
signals.
We will see these in next class.
Open area – tough problems – but
opportunities.
Copyright: S.
Krishnamurthy, UCR