On Optimizing Backoff Counter Reservation and Classifying Stations for the IEEE Distributed Wireless LANs

IEEE (MAC) DCF - Distributed Coordination Function PCF - Point Coordination Function CSMA/CA with binary exponential backoff

DCF Enhancements Many enhancements to improve performance, models with hiddent terminal Baldwin –Transmission deadline –Stations next backoff value - Enhanced Collision Avoidance (ECA) –Decrease # collisions under constant backoff window size

Proposal These studies show: Increase # competing stations --> performance sharp decrease BCR-CS Backoff Counter Reservation and Classifying Stations –Reduce collisions –Improve performance

BCR-CS Main reason for collision in DCF is that other station do not know other station’s info such as backoff counter If known, unecessary collisions and wasted waiting time can be avoided

BCR-CS Backoff counters of next frames generated in advance and sent in frame transmissions. Random backoff counter generated and embedded into header info of next frame Classify stations into 3 groups –Idle - no frame to transmit –Reserved - frames ready and backoff counters announced success through previous frames –Contentious - frames ready and not success announced

BCR-CS Frames in reserved group do not collide Frames in Contentious group do collide because of unknown backoff counter BCR-CS subschemes - based upon # stations in contentious group –BCR-CS-b - original binary backoff –BCR-CS-p - psuedo-p-persistent

Backoff Counter Table Store other station’s backoff counter’s inside table.

BCR-CS-p (pseudo-p- persistent) Goal is to avoid choosing conflicted slots already reserved by other stations Contentious Group 1) Choose smallest available backoff counter 2) Xmit when counter reaches zero 3)If frame xmit fails, repeat 1 & 2 Reserved Group –CW values are doubled if there is a collision –Collisions only possible if there are hidden nodes –Initial window size = N R + N C

BCR-CS-b (exponential backoff) Ordinary Exponential backoff Initial window size = N R + N C Reserved Group –Since collisions are very rare, use min CW –Throughput maximized when CW = 1 –Min CW may cause starvation for contentious group

Estimation of N R + N C N R is approx the # of BCT values != -1 N I is approx the # of BCT values = -1 N R + N C + N I varies as nodes move and power down We can analyze historic N C

Estimate by Time (EBT) Modify MAC headers to include time spent by the transmitting station in each state –T R (j) + T C (j) + T I (j) = 1 Delete stations from BCT have CW = -1 for long periods bc they have moved away Approximate N C by summing T C (j) Exponential smoothing can improve estimate

Estimate by Probability (EBP) Modify MAC headers to include probability that the transmitting station is in each state –Reserved frames / total frames Sum probabilities Exponential smoothing can improve estimate

Utilization vs p

Optimal p vs M (# of stations)

Optimal U vs M (# of stations)

Effects of Estimating M

Performance evaluation under two classes of traffic comprehensive evaluation for the proposed schemes comparison with the DCF and the ECA

frame payload : 500 bytes beacon interval : 100 ms DIFS time : 34 s SIFS time : 16 s slot time : 9s physical preamble : 16 s physical header time : 4s symbol time : 4 s control rate : 24Mbps data rate : 54Mbps backoff minimal window size :32 maximum backoff window size : 1,024

type A station : always has at least a frame ready to send in the queue at any time type B station : a frame only arrives after the previous frame is just transmitted NA : number of type A stations NB : number of type B stations

Pseudo-p-Persistent a)Throughput with different p values )No of collisions with different p values

Comparison over Simulation Time c) Throughput versus simulation time d) No of collisions versus simulation time

Comparison over traffic pattern a) Total throughput b) Number of collisions

Comparison over traffic pattern c) idle time d) collision time

PERFORMANCE EVALUATION UNDER q new metric q - defined as the probability that an outgoing frame arrives when the queue is not empty in a station We study performance of the proposed schemes on different traffic situations performance over the mean performance over the variance Two extreme cases of q

E(q) = 0 E(q) = 1

Comparison of Schemes under q

Effects of q Distributions

Comparison of NC Estimated Methods and Real Value

Conclusion New scheme for contention based protocol : BCR-CS Two different back off schemes Three key aspects- reservation, classification, and optimality scheme outperforms the DCF and ECA Two estimation methods of the number of contentious stations are proposed Simulation studies are performed to compare the new protocol with the DCF and ECA