Alleviating MAC Layer SelfContention in Adhoc Networks

1
Alleviating MAC Layer Self-Contention in Ad-hoc
Networks

• Zhenqiang Ye, Dan Berger, Prasun Sinha ,
• Srikanth Krishnamurthy, Michalis Faloutsos,
  Satish K. Tripathi
• Dept. of CSE, UC Riverside
• Dept of CIS, Ohio State University

2
Motivation
Self-contention Contention between packets of
same transport connection
Inter-stream contention Contention between DATA
packet stream and ACK packet stream
Intra-stream contention Contention caused by
packets of the same stream at different nodes
DATA stream (TCP or UDP)
TCP DATA stream
destination
source
source
destination
ACK stream
Contention for shared media
Contention for shared media
prior MAC solutions (none)
prior MAC solutions Fu et. al., Infocom 03

• Self-contention is best resolved at the MAC
  layer because
• Self-contention arises in the MAC layer
• Requires no changes to widely deployed
  transport protocols
• IEEE 802.11 is an evolving standard and is
  amenable to changes

3
The Main Contributions

• Propose two mechanisms
• Fast Forward alleviates intra-stream contention
  by withholding transmission until previous packet
  has reached beyond interference range.
• Quick Exchange alleviates inter-stream contention
  by exchanging TCP data and TCP ACK packets in the
  same RTS-CTS-ACK dialogue.
• Observe significant performance improvement
• Up to 250 goodput improvement
• Up to 19 backoff time reduction
• 22 MAC layer overhead reduction

4
Fast-Forward (FF)Key Idea dont send next
packet till previous packet is out of
interference range
Sender
Receiver
Next hop Receiver
RTS

• Lower avg back-off time per packet
• No backoff precedes FFPKT tx
• Fewer False Link Failures
• No explicit contention for FFPKT
• Reduced control packet overhead
• No RTS for FFPKT

CTS
DATA
Time
ACK (with Implicit RTS)
ACK( with implicit RTS)
CTS
DATA (fast forwarded packet)
FFPKT Fast Forwarded Packet
ACK( with implicit RTS)
2
6
4
6
6
2
Bytes
Frame Control
Duration
Destination Address
FCS
Modified ACK (with RTS for next-hop)
RTS dest Address
Source Address
Needed by the next hop to respond with CTS
Identifies the intended RTS recepient (next hop)
5
Quick-Exchange (QE)Key Idea subsume contention
caused by reverse stream
Receiver
Sender
Receivers Neighbor
Senders Neighbor
RTS

• Lower avg back-off time per packet
• No backoff precedes DATA2 tx
• Fewer False Link Failures
• No explicit contention for DATA2
• Reduced control packet overhead
• No RTS/CTS for DATA2
• Piggybacked ACK1

NAV (RTS)
CTS
Time
DATA1
NAV (CTS)
ACK1
DATA2
NAV (DATA1)
NAV (ACK1)
?
ACK2
2
2
2
6
4
Bytes
HCS Header Check Sequence FCS Frame
Check Sequence BSSID Basic Service Set ID
(unique network ID)
Frame Control
Duration
Extra Duration(?)
Destination Address
FCS
CTS
2
2
6
4
6
6
2
4
0 - 2308
Bytes
DATA2 (with ACK1)
Frame Control
Duration
Destination Address
HCS
Source Address
BSSID
Sequence Control
Body
FCS
ACK Header
MAC Header
6
Performance Goodput in String Topology
Single UDP flow in a string topology
Single TCP flow in a string topology
Goodput increase up to 250
Goodput increase up to 45
7
Performance Normalized Goodput in Random Topology

• 100 nodes in 2500m ? 1000m
• Average of 50 scenarios
• 180 sec sim time per scenario

TCP flows in a random topology
Normalized Goodput increases by up to 30
8
Goodput Improvement Factors(Scenario TCP flows
in random topology)
Normalized Backoff Time (backoff time per MAC
packet tx)
Normalized Control Packet Overhead (Control
packets per unicast packet)
Reduction by up to 19
Reduction from 3.2 to 2.5 (approx.)
Number of Link Failures
Reduction by up to 66
9
Conclusions

• Quick-Exchange alleviates inter-stream
  self-contention
• Fast-Forward alleviates intra-stream
  self-contention
• UDP goodput improves by 250 in string topology
• TCP goodput improves by 45 in string topology

Ongoing Work

• Goodput studies for scenarios with mobility
• Analytical model of goodput gains for
• fast-forward and quick-exchange