Title: Dynamic Adaption of DCF and PCF mode of IEEE 802.11 WLAN
1Dynamic Adaption of DCF and PCF mode of IEEE
802.11 WLAN
MTech Dissertation
- Abhishek Goliya
- 01329012
- Guided By
- Prof. Sridhar Iyer
- Dr. Leena-Chandran Wadia
15th January 2003
2Aim
- To optimize overall performance of IEEE 802.11
MAC in WLAN - Devise protocols that dynamically adapt DCF and
PCF mode of 802.11 Wireless LAN
3Road Map
- IEEE 802.11MAC
- Problem Analysis
- Suggested Protocols
- DSP Dynamic Switching Protocol
- PRRS Priority Round Robin Scheduling
- CFP Adaption
- Conclusion Future Research
4 DCF operation
5PCF Mode
6Problem Area
- Polling overhead in PCF due to unsuccessful
polling attempts - Need for Dynamic Switching between PCF and DCF to
exploit better features of both - Need of Dynamic adaptation of configuration
parameters both of PCF and DCF
7Polling Overheads
- When most stations have pending data, sequential
polling provides ordered channel access and
reduces collisions. - But when few stations have pending data polling
mechanism become significant overhead. - Unnecessary delay for stations with data
- Results in throughput degradation.
8Theoretical Analysis and Simulation Study
- TPollFail TPoll SIFS TNull SIFS
- TPollSuccess TPoll SIFS TData SIFS TAck
SIFS
Graph shows throughput degradation due to polling
overhead
9Network Monitoring Layer
- PRRS and DSP are based on monitoring layer at PC.
- Eavesdrops packets to classify nodes as Active
and Passive
Active Node Node having high probability to
transmit, when polled Passive Node Node
having low probability to transmit, when polled
10Node List Management
List Management in CFP
List Management in CP
11(No Transcript)
12Need of Dynamic Switching
- DCF perform well till networks size is small
- Suffers from throughput degradation at high load
and in large-size networks - PCF results in larger delay in small-size
networks due to polling overheads
13Learning In DCF
- Criteria of classifying node as Active is
directly applicable - How much time node should be kept in Active List?
- Station needs to transmit, take random backoff
after DIFS - Random backoff lies between 0 to Current CWsize
14Node List Management in DCF
- CWavgCWmin p 2 CWmin plevel CWmax
- level log2CWmax -log2CWmin
Event sequence and list management in DCF
15Switching Criteria (PCF to DCF)
- Keep track of number of active and passive nodes
in BSS - Switch on the basis of number of active nodes
- Approximate traffic load by keeping track of CFP
utilization
No. of stations
16Switching Criteria (DCF to PCF)
- Keeping track of number of active nodes
- Network Utilization
- Number of free slots after DIFS
- RTS Failure
- Fail to receive CTS in response to RTS
- Number of Collision
- Assuming center position for PC. It can hear
collision and approximate contention in network
17Restricted DSP
- Need fixed central coordinator (PC)
- Station need to associate with PC
- They need to follow same rules for association,
deassociation, etc. as in PCF - Network starts in PCF
- Switching decision are made at the time of beacon
transmissions
18PCF to DCF switching
- Exploits the virtual carrier sense mechanism
- Transmit beacon, as it does normally, but with
different CF parameter set - Set CFPmaxDuration, CFPremainingDur to zero and
CFPcount to non zero - PC sets its local variable cfp to zero mode to
DCF
CF parameter set
19DCF to PCF switching
- Beacon generation is centralized in DCF and done
by PC - PC waits for PIFS instead of DIFS for sending
beacons - This prioritize PC for sending beacons
- To transit into PCF mode, PC sets its CFPrate
parameter to its normal value and mode to PCF - Beacons sent by it now has actual value of
CFPmaxDuration, CFPremainingDuration and CFPcount
20Simulation Setup
- Studies confined to single cell of radius 240m
- We added support for sending Null data frame and
broadcast packet in CFP, in existing PCF patch
Throughput measured as total number of bits
passed up from the MAC sublayer at each
destination Laod measured as total number of bits
offered to the MAC sublayer at each source Delay
is measured as end to end delay at agent layer
21Simulation Result for PRRS
32 Nodes with 16 active
32 Nodes with 8 active
32 Nodes with 32 active
64 Nodes with 16 active
22Delay Graphs for PRRS
32 Nodes with 8 active
32 Nodes with 16 active
32 Nodes with 32 active
64 Nodes with 16 active
23Result Analysis
- When 25 nodes are active, throughput increases
by 10 with 32 nodes and by 15 with 64 nodes - Mean packet delay reduces when few nodes are
active - But it increases slightly when number of active
node reaches total node. - If load is moderate and few nodes are active then
PRRS gives better result
24Alternative proposals to reduce Delay further
- Use Service Differentiation mechanism in CP
- Prioritize nodes that have not been polled in CFP
by using DCF priority schemes - Use Bi-Level Feedback scheduling
- Multilevel Feedback Scheduling (MFS)
- Number of levels needed depends upon network size
- Issues regarding MFS are still open for further
research
25Simulation Results for DSP
- We vary number of active stations from 4 to 16
while simulating DSP - Both throughput and delay value improve as
compared to existing PCF and DCF
26Impact of configuration Parameter
27Effect of CFP repetition interval on PRRS
- Good value for CFP repetition interval depends
upon network size - Graph shows result with 16 and 64 nodes active
respectively from the top
16 Nodes active
64 Nodes active
28CFP Adaption Criteria
- Approximate Load by CFP utilization
- CFP utilization Time spent from first poll to
CFPend - newCFPutili. 0.8 oldCFPutili.
0.2curCFPutili. - Linear update is addition or subtraction of
beacon transmission interval - Exponential update halves the current rate
- Ensuring minimum interval and keep it as multiple
of beacon interval
C is CFP repetition count
29(No Transcript)
30Simulation Results
- CFP adapted PRRS shows better result than PRRS
with arbitrary configured CFP repetition interval
- Graph shows result with 16 and 32 active nodes
16 Nodes active
32 Nodes active
31Conclusion
- We suggested network monitoring based approach to
dynamically adapt MAC in IEEE 802.11 - PRRS improves throughput and delay values by
about 10-15 especially at moderate load and few
stations have data - DSP improves network capacity and reduces delay
under all load regime - We showed the need to dynamically adapt the
configuration parameters of both PCF and DCF - Our protocol for CFP adaption successfully adapts
CFP rate to suit current network load
32Future Work
- PRRS suffers from larger delay in some scenario
- Need to implement DCF priority schemes
- Analyse the performance of Bi-level and
multilevel feedback scheduling - Thorough experimentation and simulation study is
required to define switching points for DSP - Idea of Distributed DSP is open for further
research - Design algorithm to adapt other configuration
parameters
33