Chapter 1 Introduction
1.3 Organization
The remaining of this thesis is organized as follows. Chapter 2 describes the related work including the EDCF architecture proposed in the802.11e and its improvement. In chapter 3, the proposed media access mechanism is described in detail. And chapter 4 is the performance evaluation via simulations. The conclusion and future works is in chapter 5.
Chapter 2
Related Work
This chapter will introduce some QoS support extensions to IEEE 802.11 DCF mechanism, including official solution, and other solutions based on different adaptation aspect.
2.1 Enhanced Distributed Channel Access of IEEE 802.11e (EDCA)
IEEE 802.11e is the QoS support extension to the original IEEE 802.11 standard, and the new media access control scheme is called Hybrid Coordination Function (HCF). Because HCF is developed based on the original IEEE 802.11 standard, it also has a distributed contention-based media access function⎯Enhanced Distributed Channel Access (EDCA) (named Enhance DCF (EDCF) in early version of IEEE 802.11e draft), and a centralized contention-free media access function⎯HCF Controlled Channel Access (HCCA), which are extended from DCF and PCF, respectively. We will only focus on EDCA in this thesis because of the same reason mentioned before.
The way IEEE 802.11e provided QoS support is via service differentiation.
Unlike all the traffics are treated the same in DCF, in EDCA, all the flows are classified into four Access Category (AC), which represent the different priority and with different MAC parameters. In EDCA, all the flows use different Arbitration Inter Frame Space (AIFS[AC]), minimum contention window value (CWmin[AC]), maximum contention window value (CWmax[AC]) and Persistent Factor (PF[AC]) for
Figure 2.2-1. The timing diagram of media access in EDCF and DCF
the contention process to transmit packets belonging to the different ACs instead of original DIFS, CWmin, CWmax, PF in DCF. And the backoff timer is randomly chosen from [1, 1+CW[AC]], instead of [0, CW] as in DCF.
The timing diagram of media access in EDCA and DCF are shown in Fig. 2.2-1.
As to the definitions of these parameters belonging different ACs, in concept, the higher priority flows get the smaller values of these parameters. Because the smaller parameter values mean the higher probability to access the media, the less latency, and the more capacity share of this priority.
In EDCA, contention to the media is becoming between ACs, and each AC with a different transmission queue. Fig. 2.2-2 shows in EDCF, there are four queues in a station, where each queue behaves as a single EDCF contending entity with different parameter sets. While more than one AC within in one station attempt to transmit
Figure 2.2-2. Four access categories (ACs) in one station in EDCF
packet concurrently, the collision is handled in a virtual manner and the packet with highest priority is chosen to transmitted, which left other queues performing the contention window updating and backoff procedure. The contention window updating procedure is basically the same as in DCF, which described in follows.
1) Adjusting CW after each successful transmission
After each successful transmission, the value of contention window for class i is reset to its predefined minimum contention window value in EDCA, which described below,
[ ] min[ ]
CW i =CW i . (1) 2) Adjusting CW after each unsuccessful transmission
After each unsuccessful transmission, this class i’s contention window value become PF[i] times of previous value, of course, the new CW[i] must be bounded in the predefined value (smaller than or equal to CWmax[i]). That is,
(
max)
[ ] min [ ]* [ ], [ ]
CW i = CW i PF i CW i . (2) where in lately versions of IEEE 802.11e draft, PF[AC]s are all set to 2, which is the same as in DCF. Except for the parameters and mechanisms mentioned above, the rest part of EDCA is basically the same as DCF.
After the QoS concept is taking seriously by IEEE 802.11e, there are further QoS enhancements about improving better service differentiation, throughput or fairness.
All the further QoS enhancements can be separated into three main categories according to the aspect of adaptation: contention window based enhancement, backoff based enhancement and inter frame space (IFS) based enhancement. In the following, I am going to introduce some further QoS enhancements of different categories.
2.2 Adaptive Enhanced DCF (AEDCF) – CW-based QoS Enhancement
In the further observation of EDCA, the EDCA performance obtained are not optimal since all the MAC parameters are predefined as static values, which cannot be adapted to the network condition. Especially when the media is highly loaded, EDCA performs poorly in throughput, latency, and collision rate. This is mainly because of the immediately reducing current CW[i] to CWmin[i] after successful transmission, and leading the over high collision rate. Hence, Adaptive Enhance DCF (AEDCF) [4]
is proposed to improve EDCA by taking network condition into account in MAC scheme.
The major difference between EDCA and AEDCF is that AEDCF use the network condition to adapt CW[i], instead of setting it to CWmin[i]. The whole contention window updating procedure is shown below.
1) Adjusting CW after each successful transmission
With the motivation by the fact that when a collision occurs, a new collision is likely to occur in the near future, AEDCF adopts an approach called Slow Decrease (SD) to reduce CW by a dynamic factor. And the SD factor used here to reflect network condition is the average collision rate, which updated periodically. First, the current measured collision rate fcurrentj is calculated by the following equation:
( )
where N(collisionj[p]) is the number of collisions of node p which occurred between the (j-1)th and jth updates, and N(data_sentj[p]) is the total number of packets that have been sent by node p in the same period. It’s obvious that is always in the range of [0, 1].j current
f
Next, AEDCF uses an estimator of Exponentially Weighted Moving Average (EWMA) to smoothen the currently measured values to minimize the impact of transient collisions and get the average collision rate. The average collision rate
is calculated by the following equation.
j
where is calculated from (3), and is the measured average collision rate of the (j-1)
currentj
f faveragej−1
th update, andγ is the weight (or smoothing factor). The average collision rate is computed every period Tupdate express in time-slots, which should too long to infect the estimation preciseness and should not be too short in order to limit the complexity.
In order to ensure that the priority sequence between different ACs is still intact when a class updates its CW, each class should be assigned a different factor according to its priority level, and this factor is called Multiplication Factor (MF). In AEDCF, the maximum MF value is set to 0.8 based on set of simulations with several scenarios by the authors. And of course, the factor should not lead the calculated CW excess the previous CW because in concept, the flows transmitted successfully should not be punished more. And the MF value is determined based on (5).
[ ] min 1
( (
2 *)
averagej , 0.8)
MF i = + ×i f (5) Obviously, based on (5), the highest priority AC will reset its CW parameter with the smallest MF value. Finally, we still need to guarantee that all the computed CW after each successful transmission of packet of class i are greater than or equal to CWmin[i], so CW[i] is then updated as (7).
[ ] max
(
min[ ], [ ]* [ ])
CW i = CW i CW i MF i (7)
2) Adjusting CW after each unsuccessful transmission
After each unsuccessful transmission AEDCF did not make any change as in EDCA scheme but reset PF[i] with different value according to priority levels, which in order to re duce the probability of a new collision and consequently decrease delay.
So the CW updating equation is following the equation (2) described before. However, the PF[i] values are not set to 2 anymore, and the higher flows have the lower PF[i]
values.
2.3 Adaptive Fair Enhanced DCF (AFEDCF) – Backoff-based QoS Enhancement
While AEDCF improves the total throughput of EDCA, the performance of low-priority flows degrades sharply at high load because of the differentiation between MAC parameters of different ACs. And the fairness between the same AC and the channel utilization also degrades when the channel is congested. Hence, Adaptive Fair Enhanced DCF (AFEDCF) [5] is proposed to extend EDCA which combined the advantages of service differentiation, fast backoff decrease, and an adaptive access scheme and aim to improve (i) the performance of multimedia applications whatever is the channel load, (ii) the total throughput obtained, and (iii) the fairness between the same priority applications.
Unlike AEDCF, the contention window adjustment procedure is not the mainly part of AFEDCF and AFEDCF just follow the original mechanism of EDCA at this part. That is, after each successful transmission, the CW is updating by the equation (1) mention before; after each unsuccessful transmission, the CW is updating by the equation (2) described above. But when a queue is in deferring mode, in AFEDCF, whenever it detects the start of a new busy period (maybe caused by a collision or a
Figure 2.3-1. Backoff Timer decreasing stages in FCR mechanism
packet transmission in the media by other flows), it will react as it got through a unsuccessful transmission itself and increase the CW as above. The reason is to penalize the low priority flows and to improve the fairness between the same priority flows by having almost the same value of CW equal to CWmax[i] after the finish of a busy period, and consequently the same transmission opportunity.
The major innovation of AFEDCF is in the backoff decreasing procedure. In order to obtain better channel utilization, AFEDCF adopt a mechanism called Fast Collision Resolution (FCR) [10], and the FCR mechanism consists in using a backoff threshold value that separates two backoff states. The first backoff stage corresponds to linear decrease as in the standard. When the remaining backoff time reaches the threshold value, the queue starts the second stage by reducing the BT exponentially.
(shown in Fig. 2.3-1)
In the linear decrease stage, Backoff Timer (BT) is decreasing as following:
[ ] _ [ ], [ ] [ ]
-if BT i >Bof th i BT i =BT i SlotTime, (8) where Bof _ [ ]th i is the backoff threshold of this flow i , SlotTime is predefined system slot time. And when the remaining backoff time reaches the threshold value, the BT is decreasing exponentially until BT is zero or less than a slot time, as following:
[ ] _ [ ], [ ] [ ] / 2
if BT i ≤Bof th i BT i =BT i , (9) [ ] , [ ] 0
if BT i <ST then BT i = . (10) AFEDCF also adapted FCR by dynamically adjust the backoff threshold. In concept, when media load decreases and the queue decrements its CW[i] value, the
Figure 2.3-2. Backoff Threshold Adaptation Function
exponential decrease stage must be extended by increasing its Bof_Th[i] parameter, in order to reduce the idle time; when media load increases and the queue increments its CW[i] value, the exponential decrease stage must be reduced by decreasing its Bof_Th[i] parameter, in order to avoid a new collision. In fact, the backoff threshold function is derived by drawing a linear function (shown in Fig. 2.3-2 above) which joins the two points A(CW[i]=CWmin[i], Bof_Th[i]=CWmin[i]) and B(CW[i]=CWmax[i], Bof_Th[i]=0).
Hence, the backoff threshold adaptation function is derived as below:
max
2.4 IFS-based Distributed Fair Queuing (IDFQ) – IFS-based QoS Enhancement
Except for QoS enhancement which adapting contention window computing mechanism and backoff decreasing mechanism, there is method using inter frame space to provide better QoS support. Unlike other method working for collision resolution, IFS-based Distributed Fair Queuing (IDFQ) [6] [7] just chooses appropriate IFS values for flows with different weight, and applies some randomization to avoid collisions, all based on the concept of weighted fair queuing (WFQ). There is no backoff mechanism in IDFQ and the reason is to improve
aggregate throughput.
Since IDFQ is based on the concept of WFQ, each transmitted frame should be stamped with a finish tag which related to the weight predefined. And the frames with larger weight leads smaller finish tag, which should be transmitted before those frames with smaller weight, i.e. larger finish tag, basically. Each station i maintains a local virtual clock as a function of real time t. As in [6], in order to compatible with IEEE 802.11b MAC parameters (PIFS=30µs and DIFS=50µs), the IFS value of station i is expressed as
i( ) uniformly random number in interval [0, 10] to avoid collision, since there is no backoff mechanism. The generated IFS value will always located in interval [PIFS, DIFS], which is also proved in [6].
[ ] - ( )i F i v t
2.5 Discussion
Through the necessary simulations, as these mechanisms described above are designed in purpose, most of them reach the goal they were designed for. EDCA provides service differentiation which not provided by original DCF; AEDCF lower the collision rate and increase total throughput especially when the channel is highly load, compare to EDCA; AFEDCF performs better fairness between the same priority flows while maintaining high throughput and service differentiation; IDFQ provides higher total throughput of all flows than EDCA and service differentiation for different flows in proportion to their weights, while achieving weighted fairness between different priority flows, especially.
As to comparison under different consideration angle and different protocols, in total throughput, AEDCF, AFEDCF and IDFQ prefer better than EDCA, and AFEDCF prefers better than AEDCF. In the view of fairness, AFEDCF and IDFQ should prefer better than EDCA and AEDCF, but AFEDCF and IDFQ just achieve different kind of fairness because the method they adopted in protocols.
However, all the four mechanisms do not mention about absolute fair share of residual bandwidth among all applications, including flows of different priorities.
Even IDFQ provides only weighted fairness, i.e. relative fairness, regardless of usage or residual bandwidth. In other words, these mechanisms cannot provide global fairness. In fact, EDCA even perform better than AEDCF in this aspect.
Besides, no mechanism consider QoS demand in the aspect of transmission rate, which describes the real applications’ demand more precisely than to just define the priorities relationship. Guaranteeing that the high priority flows will get higher probability than lower priority flow may be not enough if the high priority flows demand is too high compare to the priority relation predefined, on the contrary, these
Table 2.5-1. Characteristic summation of QoS enhancement mechanisms
EDCA AEDCF AFEDCF IDFQ
Service differentiation based on priority
● ● ● ●
Service differentiation based on QoS satisfaction
○ ○ ○ ○
Total throughput improvement ○ ● ● ● Fairness between the same
priority flows
○ ○ ● ○
Weighted fairness between different priority flows
○ ○ ○ ●
Absolute fairness between all flows
○ ○ ○ ○
●: support, ○: not support
mechanism may be too unfair for the low priority flows, especially if the high priority flows’ demand are not far more the lower ones’. The characteristic analysis of mechanisms above is organized in Table 2.5-1. below.
According to the observations above, in the next chapter, I will introduce a new mechanism based on the satisfaction of applications’ transmission rate demand, and it will also achieve better global fairness among all the applications, while maintaining high total throughput.
Chapter 3
Satisfaction-based Media Access Control Scheme
In this chapter, the proposed media access control scheme is described, named Satisfaction-based Enhanced DCF (SEDCF). In the following, the description of SEDCF is separated to characteristic and assumption, parameters, and algorithm.
3.1 Characteristic and Assumption
SEDCF is capable of providing QoS guarantee for multimedia flows in the view of transmission rate satisfaction, and it ensures the global fairness among all flows while maintaining high total throughput.
In SEDCF concept, all the flows must provide their QoS demand by specifying their requirement transmission rate, not just specifying AC and get the information about priority relationships, and if the average transmission rate is higher than the required transmission rate previous defined, the flow is said to be satisfied. After a flow is satisfied, any other transmission of this flow is extra gift, regardless what AC this flow is. The concept of global fairness SEDCF provide is once the QoS flows are satisfied, the total residual bandwidth is shared fairly among all the flows, including QoS flows and best effort flows.
There are some assumptions and definitions below:
A) A node cannot transmit and receive frames simultaneously.
B) Mobility is not under consideration in SEDCF.
C) Every QoS flows must provide their QoS demand by specifying their
requirement transmission rate, not just specifying AC.
3.2 Parameters
Here are some basic parameters in SEDCF need to specify:
A) Usage
Usage means the bandwidth which s already used by a QoS flow, measure in transmission rate.
B) Minima Required transmission rate (MR)
Every QoS flow must specify MR, which represent the QoS level of this flow more precisely than just specifying AC. As to best effort flows, MR is set to be zero, that is, best effort flows are always considered to be satisfied.
C) Measuring Time Interval (Tupdate)
In every system defined Tupdate, the situations of bandwidth allocation of all flows are measured in share degree, defined below.
D) Smoothing Factor (α)
The smoothing factor is to adjust the portion of importance degree of latest estimated share degree, defined below.
E) Share Degree (SD)
In every Tupdate, Share Degree (SD) of every flow is computed. SD means how well this flow has been treated except the minima request, which also represent how much residual bandwidth this flow has used. The SD of flow i at measuring time interval j [ ] is computed by the following equation: SD ij flow i at measuring time interval j, respectively, and BW means the total j
network available bandwidth at measuring time interval j. The value is definitely between (-1, 1), and positive means the flow is satisfied at time interval j, while the negative value means the flow is not satisfied at j, which must be compensated later to ensure fairness.
j[ ]
Like measuring the network collision rate in AEDCF, in order to alleviate the impact of transient collisions, SEDCF also adopt EWMA mechanism to smoothen the estimated values. That is,
( )
1 interval j and j+1, respectively, and α is the smoothing factor here. The will be used in contention window adjustment and backoff timer decreasing procedure later.3.3 Algorithm
The SEDCF scheme is separate to two phases below: contention window adjustment and backoff timer decreasing procedure, while the detail algorithm is described in these sub-sections below.
3.3.1 Phase 1 – Contention Window Adjustment Procedure
As in EDCA, contention window needs to be adjusted only after a successful transmission or an unsuccessful transmission. Hence, the whole contention window adjusting procedure is shown as follows.
1) Adjusting CW after each successful transmission
After each successful transmission, say flow i, in the original EDCA concept, the value of contention window must be reset to CWmin[i], but in SEDCF, only the flows which are not satisfied yet ( is less than zero) have this right to do so and get more opportunity to transmit packet more, hoping for getting compensated. As to those flows which have already satisfied ( is larger than or equal to zero), basically, their contention window should be decrease slower than unsatisfied flows’
to release the transmission opportunity to other flow. Of course the decreasing potion of these satisfied flows’ CW should refer to their , CWmin[i] and CWmax[i]. Finally, the computed CW value should still be bounded between (CWmin[i], CWmax[i]), hence, the whale CW adjusting formula is derived below:
j [ ]
( )
2) Adjusting CW after each unsuccessful transmission
After each unsuccessful transmission, say flow i, on the contrary to the situation after successful transmission, as long as this flow is satisfied ( is larger than or equal to zero) now, its CW should be set to CWmax[i] to release the transmission opportunity to other flows. As to the unsatisfied flows ( is less than zero), although it should get more transmission opportunity, its CW still should increase to avoid further collision base on the basic concept of IEEE 802.11 MAC scheme. Hence, the CW of unsatisfied flows should increase slowly, and the increasing potion is computed according to their , CWmin[i] and CWmax[i]. Finally, the bounded procedure of CW is still necessary to keep CW would not be larger than CWmax[i]. The whale adjusting formula is in (16).
j [ ]
3.3.2 Phase 2 – Backoff Timer Decreasing Procedure
After the contention window is computed, if the flow i is in collision state or deferring state, the backoff timer should be randomly chosen from [1, 1+CW[i]] and start the decreasing procedure while sensing the channel is idle longer than AIFS[i],
After the contention window is computed, if the flow i is in collision state or deferring state, the backoff timer should be randomly chosen from [1, 1+CW[i]] and start the decreasing procedure while sensing the channel is idle longer than AIFS[i],