CHAPTER 3 DYNAMIC AGGREGATION SELECTION AND SCHEDULING ALGORITHM
3.2 D ETAILED O PERATIONS OF DASS
3.2.4 Fourth Phase: Scheduling packets
Future
Through the third stage, we can decide which aggregation mechanism to be adopted, and estimate for what the maximal throughput is if transmitting this kind of aggregated frames. During the second stage, under different BER conditions there will be different optimal aggregated frame size for different aggregation mechanisms, called ideal value. And comparing this ideal value with the accumulative frame size has three situations.
The first kind of situation is when the amount of frames is greater than ideal value, and then we must select enough frames from the queue to make the aggregated
size approach but smaller than ideal value. For A-MSDU, the selection strategy is First In, First Out (FIFO). However, for A-MPDU and A-PPDU, the selection strategy depends on Quality of Service (QoS) types. The frame with higher QoS type has the higher priority to be sent. If the frames are with the same QoS type, we select the frames with more hop-counts from source node to this aggregation point so that the latency between different end-to-end nodes has smaller variations. The second kind of situation is when the amount of frames is equal to ideal value. Obviously the choice is to aggregate these frames and then send out. The third kind of situation is when the amount of frames is less than ideal value. At this time, DASS will base on past traffic to predict frame arrival rate for this kind of frames. According to the past sixteen frames from now, we could estimate for frame arrival rate by taking the total frame size to divide by the time interval between the past sixteen frames. The equation for
( ) i,j
After computing frame arrival rate, we could make an estimate for whether this kind of frames will come enough to be aggregated and promote the throughput performance in the future. Below we take A-MSDU for an example. If the throughput performance by transmitting the aggregated frame made up of buffered data is defined as
Th
Buffered :
Th
Buffered= f ( D
Buffered( ) i,j , BER)
(23) Assume that we will waitT
Waiting seconds for oncoming frames in the future, the amount of frame size could be calculated by frame arrival rate:
D
Future( ) i,j = R
Predict( ) i,j * T
Waiting (24)20
Then we could deduce the equation for the throughput
Th
predict when waitingWaiting
Through the comparison between
Th
Buffered andTh
predict , we could decide whether we will wait for follow-up frames or not.
Th
predict >Th
Buffered (26) If the inequality equation above has the positive solutions, the executing step will go to main thread and hold until the arrival of the follow-up frames or the internal timeout to trigger. If the inequality equation above has no positive solutions, we will immediately aggregate all the frames in the queue and then send it out. Sometimes we determine to wait for the oncoming frames to get higher throughput, but really there are no frames that get in in the future so that makes the throughput drop off. Hence, we have to make a threshold to prevent this situation of indefinite waiting causes the throughput worse and worse. The executing step will automatically go to next step while spending more than the threshold time for waiting, but actually the throughput has decreased since waiting. At this time, the BER value will renew and the algorithm will decide the adopted aggregation mechanism again. The chosen mechanism might be not same as the former one because the quantity and the distribution of frames buffered in the transmission queue might be changed. The maximal waiting threshold is evaluated by Poisson distribution because we assume that the sequence of follow-up frames is shown as Poisson distribution. In probability theory and statistics, spending the threshold time for waiting for follow-up frames to aggregate will cause the throughput to reach the maximal performance under ideal conditions.) , ( T
k
P
λ is defined as the Poisson distribution, and the equation is :( )
Thus, for A-MSDU, the equation is expressed as:
∑
∞The computed result is namely the maximal waiting threshold.
22
Chapter 4 Simulation Results
This chapter verifies the effects of DASS through simulation by the ns-2 simulator in terms of throughput performance under infinite and steady backlog, the accuracy of prediction for frame arrival rate, and the comparisons between different selection strategies. Each scenario considers a set of algorithms supporting certain functionality. The parameters used in the simulation are shown in Table 3.
Parameter Value
Basic Rate 54 (Mbps)
Data Rate 144.44 (Mbps)
PLCP Preamble 16 (μs)
PLCP Header 48 (bits)
PLCP Rate 6 (Mbps)
MAC Header 192 (bits)
FCS (Frame Check Sequence) 32 (bits)
Time Slot 9 (μs)
Sub-frame Header in A-MSDU 14 (Bytes) Delimiter in A-MPDU 4 (Bytes) Duration of Signal Field in A-PPDU 4 (μs) RIFS (Reduced Inter Frame Space) 2 (μs) SIFS (Short Inter Frame Space) 16 (μs) DIFS (Data Inter Frame Space) 34 (μs)
Size of ACK frame 14 (Bytes)
Size of Block ACK frame 32 (Bytes)
Table 3: Simulation parameters
4.1 Simulation Environment
To test the efficiency of aggregation we assemble a noteworthy scenario that includes 16 MAPs and 10 to 30 STAs in the network. These usage models intend to support the definitions of network simulations that will allow them to evaluate performance of various proposals in terms of, for example network throughput, average latency, packet loss and other metrics. Here, we will study the maximum throughput with the proposed aggregation mechanisms when increasing the offered load with different traffic patterns. From this scenario we also observe the degrading channel efficiency when aggregation is disabled but the system is using in-full its latest PHY layer’s capabilities.
For the scenario, we set an infrastructure service area that operates in EDCA mode and includes 8 MPs and 10 to 30 STAs, all operating over a 20 MHz channel and using the same modulation coding scheme. The devices are placed over a distance of 50m and their antennas are on line of sight (LOS). The stations have the same data source that provides varying offered loads (in Mbps) of Constant Bit Rate (CBR) traffic. These CBR sources have no timeout values specified and they may have different TID. And all the data packets passed down to the MAC layer are 100Bytes in length. The BER varies from 0 to
10
−3. All simulations are run for 10 seconds.4.2 Simulation Results
4.2.1 Throughput
Throughput is obviously an important performance metric for discussing the benefit of frame aggregation. In our simulation, we designed different traffic patterns to analyze the numerical results for two topics individually. One of the topics is to
24
discuss the degree of throughput improvement under hybrid or single frame aggregation mechanisms. Thus, for this topic, the simulation was carried out with the saturated traffic and the increase of the number of STAs step by step. Figure 7 shows the throughput under the saturated traffic for frame aggregation. Comparisons with the simulation results show that the degree of the throughput improvement under hybrid adoption is apparently better than the one under single adoption. To contrast with no frame aggregation, DASS could almost promote the overall throughput for 92%. Another phenomenon we observed is that the degree of throughput improvement decreases with the increase of number of STAs. The reason is that with the increase of contentions for bandwidth the time wasted on a CSMA/CA random backoff and the probability of collisions might be raised. The situation would cause the frames to be retransmitted and make the throughput worse. There is one thing worthy to be observed is that why the throughput of the one with waiting mechanism is better than the one without waiting mechanism under saturated traffic. This is because sometimes a STA might adopt A-MSDU to aggregate the frames and then send the aggregated frame to its associated MAP, but the associated MAP might receive the aggregated frame and then consider adopting A-MPDU or A-PPDU to aggregate the received one and the buffered one into a larger size aggregated frame to court the better throughput.
Fig. 7: Frame aggregation in infinite backlog
The other topic is to discuss whether the waiting mechanism for courting better throughput performance is necessary or not. Thus, for this topic, the simulation was carried out with the unsaturated traffic and the increase of the number of STAs step by step. Figure 8 shows the unsaturated throughput for frame aggregation.
Comparisons with the simulation results show that the degree of throughput improvement with the consideration for the waiting mechanism is apparently much better than without waiting mechanism. To contrast with no frame aggregation, DASS could almost promote the overall throughput for 95%. Another phenomenon we observed is that the degree of throughput improvement increases with the increase of number of STAs. The reason should be that the total transmitted data is raised up since the channel is fully utilized and the throughput increases.
26
Fig. 8: Frame aggregation in steady backlog
4.2.2 Accuracy of Prediction of Frame Arrival Rate
In the DASS algorithm, through prediction of frame arrival rate, we can analyze and then decide whether to wait for the follow-up frames to aggregate to court better throughput. From the numerical results discussed above, for some traffic patterns under hybrid adoption in the frame aggregation mechanisms, the degree of throughput improvement with the additional waiting mechanism is further enhanced than the one without waiting mechanism. However, do the formulas in DASS for prediction of frame arrival rate determine the right time accurately? Therefore, an experiment was designed to observe the success rate, which is defined as the ratio of the times really gains better throughput to the times decides to wait, according to the playing roles in mesh. And the analysis of the success rate depends on variable number of past frames is also shown below. Figure 9 is the simulation results. From figure 9(a), the times of deciding to wait adopted by the MAPs and the MPs are much more than by STAs.
This is because the CBR traffic is generated by the STAs, the frame arrival rate of the STAs is much steady than others. Since the effect of stability, the success rate in the STAs is relatively high and approaches to 92.53%.
Except the discussion above, we also observed and analyzed the influence of changing the number of past frames used to predict frame arrival rate with exponential increase. Figure 9(a) illustrates the success rate of each aggregation point commonly drops off when the number of the referred frames increase to 128, and the degree of degradation is especially severe and evident for the STAs. We found this unusual phenomenon is caused by the CBR sources, which are off and on without stabilizing the traffic flow. If the packets generated from the STAs are transmitted continuously, with the increase of the number of the referred frames the success rate will converge and approach to a fixed value gradually. Figure 9(b) illustrates the throughput is relatively better while prediction of frame arrival rate is more precise.
Obviously, the extra waiting time caused by the failure of prediction will make the throughput abate.
Fig. 9: Accurate rate of predicting frame arrival rate
28
4.2.3 Comparisons between Different Selection Strategies
Other important issues are the frame-selection and queue-selection problems, which come up when there are many frames could be aggregated inside the queue or many queues have sufficient frames to aggregate to reach the maximum throughput at the same time. In the DASS algorithm, queue selection is to take turns between those candidates, and frame selection is to depend on the hop counts from the source to the aggregation point. A frame with more hop counts has a higher priority to be aggregated so that the deviation of access delays from their mean would be gradual.
Figure 10 shows the average latency and the throughput performance compared for the five selection strategies. The former four strategies are for frame selection, and the last one is for queue selection. The strategies for different purposes can be mixed to seek for better performance, for example, the combination of the second and the fifth.
From figure 10(a), based on the channel quality between the senders and the receivers to select the aggregated queue will decrease the average latency so that improves the throughput performance further. In order to reach the goal above, the system implemented with the multi-path scenarios is prerequisite. There is one thing worthy to be discussed is that the channel quality here is exactly the BER value measured in the second stage of DASS algorithm. Besides, the average latency we observed for different frame selection strategies varies not too much. If we analyze the variation in the average latency, it is found that the standard deviation of using FCFS is highest and the standard deviation of considering the propagation delay is lowest. This work does not discuss painstakingly limits of transmission timeout from upper layers. Users can take account of the second strategy to reduce the opportunity for timeout in reality.
Fig. 10: Comparisons between different Selection Strategies
30
Chapter 5 Conclusions and Future Works
This work aims at designing a dynamic aggregation adoption algorithm for IEEE 802.11s mesh networks in order to promote poor bandwidth utilization caused by the overhead of CSMA/CA and slow down throughput degradation caused by multi-hop transmissions.
The Dynamic Aggregation Selection and Scheduling (DASS) is proposed to achieve a high-throughput and high-efficiency mesh network. It could dynamically adopt an appropriate aggregation mechanism according to the bit error rate (BER), the communication pair, the transmission type, and the quantity and the distribution of frames in the transmission queue to maximize the bandwidth utilization as high as possible. And through the considerations above and the analysis of past traffic, we predict how many incoming frames to be aggregated, and then determine an appropriate time to send the aggregated frame.
Simulation results demonstrated that DASS algorithm actually increases the channel efficiency of the 802.11 MAC and further improves the overall throughput 95% compared with no aggregation. We have also showed that increasing PHY layer transmission rate alone does not offer higher throughputs as PHY and MAC overhead degrades the overall performance.
All types of aggregation schemes are highly recommended as they resolve the fundamental problem of existing overhead. However, the IEEE 802.11n draft only identifies the basic concepts and the data frame structures. In a flawless environment it could deliver attractive results but in terms of its functionality in a realistic environment there are still some issues that need further investigation. For example, the processing time needed to compute these mechanisms can increase the overall delays. Actually, as the efficiency of aggregation increases, its operation becomes
more complex (e.g., two-level aggregation).
Future work includes taking two-level aggregation into account and the co-existence of IEEE 802.11s draft 2.0, which is released recently and defines aggregation schemes additionally. Furthermore, mathematical modeling should be investigated to analyze the throughput performance.
32
References
[1] IEEE P802.11n/D2.0. Amendment: Medium Access Control (MAC) and Physical Layer (PHY) specifications, enhancement for higher throughput. March 2007.
[2] Yaw-Wen Kuo, “Throughput Analysis for Wireless LAN with frame aggregation under mixed traffic”, in IEEE TENCON, March 2007.
[3] D Skordoulis, Q Ni, U Ali, and M Hadjinicolaou, “Analysis of Concatenation and Packing Mechanisms in IEEE 802.11n”, in ACM Mobicom, 2003.
[4] Y Nagai, A Fujimura, Y Shirokura, Y Isota, and F Ishizu, “324Mbps WLAN Equipment with MAC Frame Aggregation for High MAC-SAP Throughput”, in JOURNAL OF COMMUNICATIONS, 2006.
[5] IEEE P802.11s™/D1.06, draft amendment to standard IEEE 802.11™: Mesh Networking. IEEE, May 2007, work in progress.
[6] Y Kim, S Choi, K Jang, and H Hwang, “Throughput Enhancement of IEEE 802.11 WLAN via Frame Aggregation”, in IEEE Technology Conference, 2004.
[7] YS Lin, JY Wang, and WS Hwang, “Scheduling Mechanism for WLAN Frame Aggregation with Priority Support”, in Vehicular Technology Conference, Fall. 2002.
[8] S Kuppa and GR Dattatreya, “Modeling and Analysis of Frame Aggregation in Unsaturated WLANs with Finite Buffer Stations”, in IEEE Communications, 2006.
[9] J Yin, X Wang, and DP Agrawal, “Optimal Packet Size in Error-prone Channel for IEEE 802.11 Distributed Coordination Function”, in IEEE Wireless Communications and Networking Conference, 2004.
[10] Y Lin and VWS Wong, “Frame Aggregation and Optimal Frame Size Adaptation for IEEE 802.11n WLANs”, in IEEE GlobeCOM, 2006.
[11] S Yun, H Kim, H Lee, and I Kang, “Improving VoIP Call Capacity of Multi-Hop Wireless Networks through Self-Controlled Frame Aggregation”, in IEEE Vehicular Technology Conference, 2006.
[12] S Kim, SJ Lee, and S Choi, “The Impact of IEEE 802.11 MAC Strategies on Multi-Hop Wireless Mesh Networks”, in Wireless Mesh Networks, 2006.
[13] R Riggio, FD Pellegrini, and N Scalabrino, “Performance of a Novel Adaptive Traffic Aggregation Scheme for Wireless Mesh Networks”, in IEEE Wireless Networks, Spring 2005.