1. Frederico Calì Marco Conti Enrico Gregori Presented by Andrew Tzakis
Dynamic Tuning of the IEEE Protocol to Achieve a Theoretical Throughput Limit
Frederico Calì
Marco Conti
Enrico Gregori
Presented by Andrew Tzakis

2. Motivation Wireless networks are slower than Wired networks
Mainly do to the use of a common medium
MAC – Coordinator of use
IEEE Currently uses Binary back off
It can be improved

3. Main Idea
Define an anylitical model in order to maximize the capacity of the network
This can be achieved through tuning of the IEEE back off algorithm
Instead of using binary back off, sample contention window (CW) from a geometric distribution.

4. Outline
Show that depending on network conditions, IEEE can be far from max capacity.
Analytical model
Model IEEE
Compare to max
An optimal constant back off can operate at close to the theoretical max
Find optimal CW
Define distributed algorithm

5. Capacity Capacity: Maximum value of bandwidth ρmax = m/tv
m = Average time busy sending a successful message
tv = Virtual Transmission Window

6. Virtual transmission length
Average virtual transmission time length E(TV) can be expressed as
virtual transmission time (Tv)
collision
successful
Ncol 1
Ncol 2 …

7. Math – Geometric Distribution
Run an experiment until a success is found
Success occurs with a probability of: p
Expected attempts before a success is 1/p

8. Virtual transmission length (2)
Since the model is based on the CW being sampled from a geometric distribution Future behavior does not depend on past we can simplify:

9. More on p
Assumption: Every node has a packet ready to send at all times (asymptotic condition)
In this case, p is the attempt probability
At the beginning of a slot this is the probability that the back off timer is equal to zero
P = 1/(E[B]+1) (property of geometric dist.)
E[B] = number of slots before a success (expected value)
p can be used to define each part of tv

10. Virtual transmission length (3)

11. Applying IEEE 802.11 to Model Need to find corresponding p value
p(i) = 2/(E[CW(i)]+1)
This means we can find E[CW] for IEEE which is based on binary back off
Approximate with linear sequence: |E[CW(n)] - E[CW(n-1)]| < ε

12. Using Calculated CW Analytic and simulated results are very close!
Model provides good approx.

13. Capacity of IEEE 802.11 ρmax is a function of p,M,q
p(i) = 2/(E[CW(i)]+1)
M = Number of nodes
q = Geometric distributed size of message
q Ranges from 0.5 – 0.99

14. Where is the room for improvement?
High number of idle slots
High number of collisions
Balanced idle slots and collations

15. Using the optimal value of p

16. IEEE To find best CW
Through monitoring the radio, the collision length and number of collisions can be found
Use minimization algorithm to find p
This is too time consuming to solve
Create an approximation
Given:
When p is low tv is determined by E[Idle_p]
When p is high tv is determined by number of collisions
Balance should be when collision time equals idle time
E[Coll]*E[Nc] = (E[Nc]+1)*(E[Idle_p])

17. Estimated values are close
Results of Estimation
Estimated values are close
To the optimal values

18. IEEE 802.11+ is much closer to the analytical bound
Capacity with IEEE
IEEE is much closer to the analytical bound

19. Distributed algorithm to get M
Since Total_Idle_p = (E[Nc]+1)E[Idle_p] Total_Idle_p = (1-p)/M*p M = (1-p)/p*Total_Idle_p
Through tracking the idle time, and out estimated p we can calculate M

20. Conclusion Derived theoretical limit
Closely approximates IEEE
Demonstrated that it is possible to tune the CW at runtime
Finding the pmin can be used in other algorithms as well (can make a hybrid)