Dynamic Tuning of the IEEE Protocol to Achieve a Theoretical Throughput Limit Frederico Calì, Marco Conti, and Enrico Gregori IEEE/ACM TRANSACTIONS ON NETWORKING, VOL. 8, NO. 6, DECEMBER 2000
Overview Introduction IEEE Capacity Analysis IEEE Protocol Comments Conclusions
Introduction – (1) The existing IEEE protocol is contention-based, which allows stations to access the wireless channel randomly. Its performance degrades under heavy network load due to the high collision probability.
Introduction – (2) Motivation: find the optimal contention window (CWIN) size which boosts up the throughput. The value of CWIN which is large enough to minimize the collision probability and small enough to minimize the backoff interval.
Introduction – (3) An analytical model is built to derive the average size of the contention window which maximizes the throughput. Observations: Depending on the network configuration, the standard can operate very far from the theoretical throughput limit. An appropriate tuning of the backoff algorithm can drive the IEEE protocol close to the theoretical throughput limit.
Introduction – (4) Hence a distributed algorithm IEEE is proposed that enables each station to tune its backoff algorithm at run-time.
IEEE Capacity Analysis (1) No RTS/CTS mechanism No dependency on the physical layer technology WLAN configuration
A model with a finite number of stations (M) which always have a packet ready for transmission. For each transmission attempt, a station is assumed to use a backoff interval sampled from a geometric distribution with parameter p, where IEEE Capacity Analysis (2)
In the real IEEE backoff algorithm, a station transmission probability depends on the history. However, our model of the protocol behavior provides accurate estimates of the IEEE protocol behavior from a capacity analysis standpoint. IEEE Capacity Analysis (3)
Protocol Capacity (Max throughput) t v = average virtual transmission time m - = average message length IEEE Capacity Analysis (4) ……..(1)
S = time required to complete a successful transmission (DATA-ACK frame exchange without collision) IEEE Capacity Analysis (5)
Derivation of the average message length IEEE Capacity Analysis (6)
IEEE Capacity Analysis (7) Derivation of E[B]: Corollary: Given Lemma 2
Derivation of the average t v IEEE Capacity Analysis (8) ……..(2)
The assumption on the backoff algorithm implies that the future behavior of a station does not depend on the past. The idle period times are i.i.d. sampled from a geometric distribution with an average. The collision lengths are i.i.d with average. IEEE Capacity Analysis (9)
IEEE Capacity Analysis (10) ……..(3)
IEEE Capacity Analysis (11)
From (1), (3) and Lemma 3 IEEE Capacity Analysis (12) ……..(4)
Validate the analytical model via simulation. Given M and q, how close are the throughput derived analytically and that derived from simulation ? IEEE Capacity Analysis (13)
IEEE Capacity Analysis (14)
IEEE Capacity Analysis (15)
Given M and q, find p which maximizes the capacity in (4). M = the number of stations p = a parameter for the geometric distribution of the backoff time q = a parameter for the geometric distribution of the packet length (number of slots) IEEE Capacity Analysis (16)
IEEE Capacity Analysis (17) For a given packet length distribution, the maximum value of the capacity corresponds to the minimum value of the average virtual transmission time (t v ).
IEEE Capacity Analysis (18) Goal: Find p (p min ) to minimize tv:
IEEE Capacity Analysis (19)
IEEE Capacity Analysis (20)
Motivation Based on the simulation, the IEEE protocol with an appropriate setting of the CWIN size (optimal CWIN) can reach the theoretical limit. IEEE (1)
IEEE (2)
Tune the value of E[CWIN] based on p min at run time such that the capacity is maximized. Both E[Coll] and E[N c ] can be estimated by observing the channel status. A station can obtain p min with a minimization algorithm. IEEE (3)
The minimization algorithm is NOT suitable for a run-time computation as it very complex computationally. A heuristic is introduced to approximate p min. IEEE (4)
Approximate p min by finding p satisfying The selection of p min guarantees that E[Coll] < 1. IEEE (5)
IEEE (6)
IEEE (7)
The contention window size is updated at the end of any virtual transmission time that contains at least one collision. IEEE (8)
IEEE (9)
The previous results were obtained under the following assumptions: M (the number of stations) is known before. No hidden terminals Study how IEEE is sensitive to the number of active stations Hidden terminals IEEE (10)
Knowing M beforehand is a strong assumption. In a real network, the number of active stations is highly variable. M cannot be more or less the same all the time Relax this assumption and analyze the sensitiveness of the IEEE capacity to the number of active stations. IEEE (11)
IEEE (12) Assume M=100
The number of active stations (M) should be estimated at run time. M can be computed provided that the average number of empty slots in a virtual transmission time is known. IEEE (13)
IEEE (14)
To avoid sharp changes in the estimated value of M, is introduced IEEE (15)
IEEE (16)
IEEE (17) 10010
IEEE (18)
Relax the 2 nd assumption – Hidden stations exist. Study how hidden stations affect the performance of our protocol by causing erroneous statistics. Analyze the impact on our protocol of three events that may occur when hidden stations are present. missed ACK, carrier sensing fault and not-detected transmission IEEE (19) – Hidden stations
Missed ACK The hidden station problem may cause a station to miss the ACK. Consequences Observes a longer virtual transmission time Considers a successful transmission to be a collision IEEE (20) – Hidden stations
IEEE (21) – Hidden stations Missed ACK
Carrier Sensing Fault A station (1) wrongly senses the wireless medium has been idle while a hidden station is transmitting and (2) starts transmitting. Consequences Generate collisions IEEE (22) – Hidden stations
Non-detected Transmission A station cannot observe some collisions and successful transmissions and mistakenly interpret that the channel is idle. Consequences Overestimate the idle-period length. Underestimate the number of collision and the collision length. IEEE (23) – Hidden stations
A probabilistic model is used to associate each phenomenon to a probability H1: the probability that a station misses an ACK. H2: the probability that, due to a carrier sensing fault, a station which does not detect an ongoing transmission starts transmitting. H3: the probability that a station overestimate the idle-period length. IEEE (24) – Hidden stations
IEEE (25) – Hidden stations (50 active stations) Given H1=H2=H3=H
IEEE (26) – Hidden stations
IEEE (27) – Hidden stations
IEEE (28) – Hidden stations
Good Make valid assumptions - the geometric distribution of backoff time E[B]. Justified by comparing figures derived from the simulation and the mathematical analysis. Demonstrate how to handle the big assumption step by step. Comments – (1)
Bad The overhead incurred for estimating p min is not shown. It won’t appear if IEEE is simulated in ns-2. Comments – (2)
Conclusions A distributed algorithm IEEE is proposed that enables each station to tune its backoff algorithm at run-time. The throughput can be optimized such that it is close to the theoretical upper bound.