Chapter 3 Satisfaction-Based Medium Access Control Scheme
3.3 Algorithm
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], and the flow cannot attempt to transmit packet only after the backoff timer is decreasing to zero. Unlike AEDCF, in order to maintain the global fairness between all flows, even in backoff timer decreasing procedure, those unsatisfied flows ( is less than zero) should decrease their BT[i] faster to zero, which can make more opportunity to transmit next time. In SEDCF, the FCR mechanism is used on unsatisfied flows, which decreasing BT[i] exponentially, that is
j [ ]
average
SD i
(
averagej [ ] 0 , [ ])
[ ] / 2if SD i < BT i =BT i . (17)
As to satisfied flows ( is larger than or equal to zero), their BT[i]
decrease slowly than unsatisfied flows do, which the reason is that they have already get their minima request and should release the media to other flows. Hence their BT[i]
still decrease linearly as the EDCA mechanism, which the formula is
j [ ]
average
SD i
(
averagej [ ] 0 , [ ])
[ ]if SD i ≥ BT i =BT i −SlotTime. (18)
Chapter 4
Performance Evaluation
In this chapter, the performance evaluations of SEDCF, AEDCF and AFEDCF will be proposed by using ns-2 simulator [11].
4.1 Simulation Environment
Besides using ns-2 simulator, other simulation environment is described as follows. IEEE 802.11a is adopted as the PHY layer, and detailed parameters are listed in Table 4.1-1, including total data rate, Slot_time, which is significant to the proposed mechanism.
Because SEDCF, AEDCF and AFEDCF are proposed based on EDCA of IEEE 802.11e, all the parameters used in IEEE 802.11e MAC layer to provide service differentiation are in Table 4.1-2 for the general simulations later. We generate three classes of traffic in our simulations, i.e., phone, video, and best effort flows, respectively. These three types of flows represent the highest, the second, and the lowest priority, accordingly. All three classes of flows send data with constant data rates of 160 bytes per 20 ms, 1280 bytes per 10 ms, and 200 bytes per 12.5 ms, respectively. The simulation time is 12 seconds. We assume that all flows are backlogged during the simulation time. We set the QoS demands for phone and video flows are both 50% transmission successful rate. That is, to be satisfied, the minima transmission rate requirement for phone and video flows are 32 Kbps and 512Kbps, respectively. Furthermore, based on [4], the smoothing factor and Tupdate is set to 0.8 and 5000 Slot_time, accordingly.
In order to increasing the network load, the number of nodes will increase
gradually to simulation. All the nodes locate in the same Basic Service Set (BSS), and the diagram of the traffic is shown in Fig. 4.1-1, which is that every node sends three distinct flows to next node, and all the traffics are one-hop.
Table 4.1-1. Parameter settings of PHY layer
SIFS 16μs Preamble Length 20μs
RxTxTurnaround time 1μs PLCP header length 4μs
Table 4.1-2. Parameter settings of IEEE802.11e MAC layer
Parameters Phone
Figure 4.1-1. Simulation scenario
4.2 Performance Metrics
The performance metrics measured in the simulation include the network throughput, satisfaction index, and fairness index, which extended from [12] as defined below:
A) Network throughput (ϕ)
The summation of all flows’ Usage, i.e., ( ),
i
Usage i i F
ϕ=∑ ∀ ∈ , (19) where F is the set of all flows, Usage i( ) is the usage of flow i.
B) Satisfaction index (η)
It only counts for QoS flows and is used to indicate the satisfaction degree. Its definition is
MR i are the usage and the minima required transmission rate of flow i.
The concept is, say flow i, once , it is said satisfied, for satisfaction index, how much the media is over used by this flow is meaningless, so we do not have to consider to compute
[ ] [ ] Uasge i ≥MR i
( )
Usage i x ; while for the unsatisfied i flows, how much more the usage needs for them to satisfy is very important, for satisfaction index, and the closer x to 1, the closer this flow is satisfied. And i the final value of η is between 0 and 1 after indexing normalization. The larger the value η is, the better the overall satisfaction degree of QoS flows is.
C) Fairness index (κ)
It counts all flows and is to show how fair share about the residual bandwidth. Its definition is
MR i are the usage and the minima required transmission rate of flow i. On the contrary concept of satisfaction index, for any unsatisfied flow, the difference between its usage and its minima required transmission rate is not important, because fairness index is about residual bandwidth. As to satisfied flows, how much usage a flow over used is very significant, and the fairness we attempt is among all flows regardless of priority, so yi is not concern about MR i in [ ] denominator. The larger difference between all flows’ leads the worse fair share among all flows. After the indexing normalization, the final value of κ is also between 0 and 1. The larger the value κ is, the more fairly share of the residual bandwidth among all flows.
yi
D) Mean Delay (δ)
The mean end-to-end delay is the time difference of a QoS packet from source to destination, i. e.,
_ ( ), Mean Delay i i F
δ = ∀ ∈ (22) where F is the set of all flows, Mean Delay i is the average value of all the _ ( ) end-to-end delay of flow i.
4.3 Simulation Result
In order to understand the performance of SEDCF precisely, the simulation results will be apart to phase by phase. That is, in the following sub-section, I will propose the baseline comparison of those related works, performance comparison of SEDCF phase 1 vs. AEDCF, and then SEDCF phase 1 vs. SEDCF phase 1+2, SEDCF phase 1 vs. AFEDCF, the delay comparison, finally is SEDCF phase 1 vs. AFEDCF.
4.3.1 Baseline Comparison of Related Works
First of all, we propose the baseline comparison of related works, which is include EDCA, AEDCF and AFEDCF, and IDFQ is not included because it is based on WFQ, which is totally different concept from others.
Fig. 4.3-1 shows the throughput of EDCA, AEDCF and AFEDCF. We can see that the throughput lines increase before there are 15 nodes, and decrease after that, because after there are 15 nodes, the total available bandwidth is not enough to handle
0 500 1000 1500 2000 2500
5 10 15 20 25 30 35 40
Number of nodes
Overall throughput (KB/s)
EDCA AEDCF AFEDCF
Figure 4.3-1. Overall throughput of EDCA, AEDCF and AFEDCF
0.7 0.75 0.8 0.85 0.9 0.95 1 1.05
5 10 15 20 25 30 35 40
Number of nodes
Overall satisfaction index
EDCA AEDCF AFEDCF
Figure 4.3-2. Overall satisfaction index of EDCA, AEDCF and AFEDCF
all the traffic. After there are 15 nodes, AFEDCF performs outstandingly in the related works.
There are overall satisfaction index and overall fairness index comparison shown in Fig. 4.3-2 and Fig 4.3-3. In the overall satisfaction index, it includes all the QoS flows, which mean it does not include best effort flows. All the satisfaction indexes start to degrade after there are 20 nodes, and EDCA and AEDCF have no big difference while AFEDCF is the outstanding method (over 0.9 even when there are 40 nodes) again.
0 0.2 0.4 0.6 0.8 1
5 10 15 20 25 30 35 40
Number of nodes
Overall fairness index
EDCA AEDCF AFEDCF
Figure 4.3-3. Overall fairness index of EDCA, AEDCF and AFEDCF
As to overall fairness index, after there are 15 nodes, AEDCF performs worst in three protocols, while EDCA and AFEDCF performs overall satisfaction index over 0.4. AFEDCF performs satisfaction index about 0.6 by indirectly achieving inter class fairness after there are 30 nodes.
4.3.2 SEDCF phase 1 vs. AEDCF
Since SEDCF phase 1 and AEDCF are similar to adjust contention window by a periodically estimated factor, and neither adapt the original backoff timer decreasing procedure, we propose their performance comparison first.
The throughputs of SEDCF phase 1 and AEDCF are shown in Fig. 4.3-4. We found that SEDCF phase 1 has better video-type flow and overall throughput than that of AEDCF. The reason is we lower the sending failure rate of QoS flows by adjusting the CW of satisfied flows more flexibly. And the throughout of phone–type flow is maintained the same as that of AEDCF. Furthermore, as the number of nodes
0 500 1000 1500 2000 2500
5 10 15 20 25 30 35 40
Number of nodes
Throughput (KB/s)
Phone-SEDCF p1 Video-SEDCF p1 BE-SEDCF p1 Overall-SEDCF p1 Phone-AEDCF Video-AEDCF BE-AEDCF Overall-AEDCF
Figure 4.3-4. Throughput of SEDCF phase 1 and AEDCF
increasing, the throughput of phone-type flows keeps increasing; contrarily, the throughput of video-type flows starts to decreasing when the number of nodes is larger than 15. The reason is that in such a case that 15 nodes are backlogged to send data, the total required bandwidth to satisfy their QoS demands almost equals to the available bandwidth. Thus, more number of nodes, more number of the highest-priority flows (i.e., phone-type flows). In such situation, to guarantee phone-type flows’ QoS demands, best effort-type and even video-type flows should sacrifice to release some bandwidth.
0.7 0.75 0.8 0.85 0.9 0.95 1 1.05
5 10 15 20 25 30 35 40
Number of nodes
Satisfaction index
Phone-SEDCF p1 Video-SEDCF p1 Overall-SEDCF p1 Phone-AEDCF Video-AEDCF Overall-AEDCF
Figure 4.3-5. Satisfaction index of SEDCF phase 1 and AEDCF
Fig. 4.3-5 shows the satisfaction index of SEDCF phase 1 and AEDCF. We found both phone-type traffics of SEDCF phase 1 and AEDCF have same high value satisfaction index, however, the other satisfaction index of both SEDCF phase 1 and AEDCF are slightly decreasing while the number of nodes increase because the available bandwidth is no longer enough to satisfy the QoS demand of those Video-type flows. But because we take account of SD into CW adjustment, most of the flow satisfaction index of SEDCF phase 1 aggregate better than those of AEDCF, which leads the higher overall and video-type flow satisfaction index.
0
Figure 4.3-6. Fairness index of SEDCF phase 1 and AEDCF
The measured fairness index is shown in Fig. 4.3-6 Similar to satisfaction index, flows of phone-type have the best fairness index (more than 0.98) than others. As to the other flows of AEDCF, the fairness index is decreasing distinctly after the number of node is more than 15. While the other flows of SEDCF phase 1 have generally constant fairness indexes, which result from taking SD into account in adjusting CW provides well intra-class (local) and inter-class (global) fairness. But there is an exception while there are 25 nodes in topology, at this time, the available bandwidth can just no longer provide the video-type QoS demand (529.99 kbits/s per flow, which is very close to the require transmission rate 512 kbits/s per flow), which leads to the residual bandwidth of video-type flows distributed separately, and the fairness index is lower. But while the number of nodes keeps growing, the residual bandwidth of video-type flows aggregated soon although the QoS demand is no longer satisfied, so the fairness indexes afterward go back to higher value.
0
Figure 4.3-7. Throughput of SEDCF phase 1 and SEDCF phase 1+2
4.3.3 SEDCF phase 1 vs. SEDCF phase 1+2
After tuning of the contention window, the performance of adding the new backoff timer decreasing procedure should be evaluated. Fig. 4.3-7 shows the throughput comparison of SEDCF phase 1 and SEDCF phase 1+2. The phone-type flow throughput is still increasing gradually and stably while the number of node increase. As to video-type flows, after there are 20 nodes, the throughput of SEDCF phase 1+2 video-type flows start decreasing because of the total available bandwidth is running out for the total QoS demand of QoS flows, which makes the total throughput of SEDCF phase 1+2 reached high peak about 2200 KB/s, even higher than SEDCF phase 1 at all time. The reason is for the unsatisfied flows, SEDCF phase 1+2 provide even better protection by counting their backoff timer faster than satisfied flows, and since the best effort are always considered as satisfied, they can never benefited from the mechanism and start sacrifice to maintain QoS flows demand
0.7
Figure 4.3-8. Satisfaction index of SEDCF phase 1 and SEDCF phase 1+2
earlier, which makes the highest throughput ever.
The satisfaction index of SEDCF phase 1 and SEDCF phase 1+2 are shown in Fig. 4.3-8. SEDCF phase 1+2 also performs well at this part. The satisfaction indexes of phone-type and video-type flows are close to 1, although there are lightly degrade as the number of node increase, they are never lower than 0.98. And the overall satisfaction index is never lower than 0.9 even when there are 40 nodes in the network.
The reason is SEDCF provide almost perfect protection to QoS flows by taking SD into account to compute CW and decreasing BT.
Fig. 4.3-9 shows the fairness index of SEDCF phase 1 and SEDCF phase 1+2.
For fairness between flows of the same priority, SEDCF phase 1 and SEDCF phase 1+2 almost performs the same, and the fairness indexes are almost 1 at all time except when there are 25 nodes in SEDCF phase 1 and there are 30 nodes in SEDCF phase 1+2. The fairness index drop reason of SEDCF phase 1+2 is the same as SEDCF phase1: at this time, the available bandwidth can just no longer provide the video-type
0 0.2 0.4 0.6 0.8 1 1.2
5 10 15 20 25 30 35 40
Number of nodes
Fairness index
Phone-SEDCF p1 Video-SEDCF p1 BE-SEDCF p1 Overall-SEDCF p1 Phone-SEDCF p1+2 Video-SEDCF p1+2 BE-SEDCF p1+2 Overall-SEDCF p1+2
Figure 4.3-9. Fairness index of SEDCF phase 1 and SEDCF phase 1+2
QoS demand, which leads to the residual bandwidth of video-type flows distributed separately. However, the cause leads in the fairness index drop time difference is the QoS flows protection again: SEDCF phase 1+2 take SD into backoff timer decreasing procedure, and extend the video-type flow satisfied life to about there are 30 nodes (462.05 kbits/s per flow, which is very close to the require transmission rate 512 kbits/s per flow). For the overall fairness index, since best effort flows can not benefit from adding new backoff timer decreasing procedure and their CWmin and CWmax and other MAC parameters are week to get media access compare to QoS flows, the overall fairness index of SEDCF phase 1+2 is lower than SEDCF phase 1, even the intra-class fairness of best effort flows in SEDCF phase 1+2 is maintained higher than 0.95 at all time.
0 500 1000 1500 2000 2500
5 10 15 20 25 30 35 40
Number of nodes
Throughput (KB/s)
Phone-SEDCF p1 Video-SEDCF p1 BE-SEDCF p1 Overall-SEDCF p1 Phone-AFEDCF Video-AFEDCF BE-AFEDCF Overall-AFEDCF
Figure 4.3-10. Throughput of SEDCF phase 1 vs. AFEDCF
4.3.4 SEDCF phase 1 vs. AFEDCF
In section 4.3.1, generally speaking, AFEDCF performs best in the view of throughput and satisfaction index, even in the view of fairness index. In section 4.3.3, SEDCF phase 1+2 achieves higher throughput but lower fairness index than SEDCF phase 1 does. Hence, in this section, we are going to exam the performance of SEDCF phase 1 and AFEDCF.
Fig. 4.3-10 shows the throughput of SEDCF phase 1 and AFEDCF. The overall trend of these results is the same as above, which also means the overall throughput is decreasing after there are 15 nodes, while the throughput of phone-type flow increases steady. The difference between the overall throughput of SEDCF phase 1 and AFEDCF is not really large, which means SEDCF phase 1+2 will achieve higher throughput than AFEDCF does. In basic, the throughput performance of SEDCF phase 1 and AFEDCF is similar.
0.7 0.75 0.8 0.85 0.9 0.95 1 1.05
5 10 15 20 25 30 35 40
Number of nodes
Satisfaction index
Phone-SEDCF p1 Video-SEDCF p1 Overall-SEDCF p1 Phone-AFEDCF Video-AFEDCF Overall-AFEDCF
Figure 4.3-11. Satisfaction index of SEDCF phase 1 vs. AFEDCF
Similar as throughput performance, the difference between overall satisfaction indexes of SEDCF phase 1 and AFEDCF is not large. But as we can see in Fig. 4.3-11, the video-type is better protected by SEDCF phase 1, because SEDCF phase 1 take minima required transmission rate to adjust CW, while AFEDCF just provides priority-based QoS support to QoS flows, which may lead lower QoS flows (video-type flows) may sacrifice sooner under the consideration of required transmission rate.
0 0.2 0.4 0.6 0.8 1 1.2
5 10 15 20 25 30 35 40
Number of nodes
Fairness index
Phone-SEDCF p1 Video-SEDCF p1 BE-SEDCF p1 Overall-SEDCF p1 Phone-AFEDCF Video-AFEDCF BE-AFEDCF Overall-AFEDCF
Figure 4.3-12. Fairness index of SEDCF phase 1 vs. AFEDCF
The fairness index of SEDCF phase 1 and AFEDCF is shown in Fig. 4.3-12. The performance of SEDCF phase 1 and AFEDCF are not different until there are more than 15 nodes. The fairness indexes of phone-type flows and best effort flows under two protocols are all over 0.9 no matter how many nodes are there. After there are 15 nodes, both the fairness indexes of video-type flows in SEDCF phase 1 and AFEDCF degrade sharply because the total bandwidth is running out, but fairness index of video-type flows in SEDCF phase 1 reach back to high value sooner (after there are 25 nodes), while the same situation happens in AFEDCF while there are 40 nodes.
This represents SEDCF phase 1 provides better intra class fairness between video-type flows. As to overall fairness index, the performance of SEDCF phase 1 and AFEDCF are almost on a par, the two protocols both provide over certain degree of inter class fairness.
0 500 1000 1500 2000 2500
5 10 15 20 25 30 35 40
Number of nodes
Mean delay (ms)
Phone-AEDCF Video-AEDCF Phone-AFEDCF Video-AFEDCF Phone-SEDCF p1 Video-SEDCF p1 Phone-SEDCF p1+p2 Video-SEDCF p1+p2
Figure 4.3-13. Mean delay of AEDCF vs. AFEDCF vs. SEDCF phase 1 vs.
SEDCF phase 1+2
4.3.5 Mean delay of AEDCF vs. AFEDCF vs. SEDCF phase 1 vs.
SEDCF phase 1+2
Here this section illustrates the comparison of mean end-to-end delay between AEDCF vs. AFEDCF vs. SEDCF phase 1 vs. SEDCF phase 1+2. As satisfaction index, mean delay is also calculated for QoS flows. As we can see in Fig. 4.3-13, the delay of phone-type flow is always bounded in certain area, even in the traffic load is high, which shows that the high priority flows is protected well no matter what protocol is adopted. As to video-type flows, after there are 15 nodes, the delay increase more sharply than that of flow flows because the total available bandwidth is running out, generally speaking, SEDCF performs better than AEDCF and AFEDCF and the difference is getting larger while the number of nodes is increasing, although SEDCF phase1 and SEDCF phase 1+2 are not designed for controlling delay.
Table 4.3-1. Parameter settings of IEEE802.11e MAC layer
4.3.6 SEDCF phase 1 vs. SEDCF phase 1+2 vs. AFEDCF
In order to investigate SEDCF’s performance of QoS guarantee more detail under admission control. This special scenario is upon the same ring topology and assumes at there are just 10 nodes in the ad hoc network to definitely be sure that each QoS flow’s minimum demand can be guaranteed. The MAC parameters used in this scenario are listed in Table 4.3-1. There are still three flow priorities, and all are with the same MAC parameters, constant sending rate and same packet size to eliminate the defect of best effort flows, and the setting here is also compatible to original IEEE 802.11 MAC protocol. The major difference between priorities is the minima required transmission rate. To be satisfied, the QoS demand for high priority and media priority flows are set to be 384 Kbps and 256Kbps, i.e. 75% and 50% successful transmission rate, respectively. Furthermore, the smoothing factor and Tupdate is still 0.8 and 5000 Slot_time, accordingly.
0 200 400 600 800 1000 1200 1400
High priority Medium priority Low priority Overall flow
Thoughput (KB/s)
AFEDCF SEDCF p1 SEDCF p1+2
Figure 4.3-14. Throughput of SEDCF phase 1 vs. SEDCF phase 1+2 vs. AFEDCF
The network throughput and fairness index of SEDCF phase 1 and SEDCF phase 1+2 and AFEDCF are in Figs. 4.3-14 and 4.3-15. We found that the throughputs of these three mechanisms have no much difference in overall throughput, while SEDCF phase 1 and SEDCF phase 1+2 provide QoS guarantee to sacrifice best effort flows.
The network throughput and fairness index of SEDCF phase 1 and SEDCF phase 1+2 and AFEDCF are in Figs. 4.3-14 and 4.3-15. We found that the throughputs of these three mechanisms have no much difference in overall throughput, while SEDCF phase 1 and SEDCF phase 1+2 provide QoS guarantee to sacrifice best effort flows.