Efficient dynamic frame aggregation in IEEE
802.11s mesh networks
Ying-Dar Lin
1,2, Jui-Hung Yeh
2,∗,†, Tsung-Hsien Yang
1, Chia-Yu Ku
1,2,
Shiao-Li Tsao
1,2and Yuan-Cheng Lai
31Department of Computer Science, National Chiao Tung University, Hsinchu, Taiwan 2Realtek-NCTU Joint Research Center, National Chiao Tung University, Hsinchu, Taiwan
3Department of Information Management, National Taiwan University of Science and Technology, Taipei, Taiwan
SUMMARY
IEEE 802.11n could improve the network efficiency of WLAN by aggregating multiple frames into a single transmission. The frame aggregation technique especially benefits multi-hop transmissions that introduce enormous 802.11 medium access control (MAC) layer control overhead. This paper proposes a novel algorithm, namely dynamic aggregation selection and scheduling(DASS) algorithm, that can dynamically adopt the appropriate aggregation mechanism for IEEE 802.11s mesh network. The characteristics, benefits and the restrictions of three aggregation mechanisms, aggregate MAC service data units, aggregate MAC protocol data unit and aggregate physical protocol data unit, are investigated. Based on the channel condition, the quantity and the distribution of the frame arrivals, DASS could determine what aggregation mechanism to adopt and whether to wait for the next frame for aggregating transmissions. Simulation results demonstrate that the algorithm significantly increases the channel efficiency of the 802.11 MAC and further improves the overall throughput of the wireless mesh network by approximately 95%. Copyright q 2009 John Wiley & Sons, Ltd.
Received 1 December 2008; Revised 7 April 2009; Accepted 9 April 2009 KEY WORDS: frame aggregation; multi-hop; mesh networks; 802.11s
1. INTRODUCTION
With the increasing demand for real-time applications over wireless networks, IEEE 802.11n[1] is proposed to provide high-speed wireless access with maximum physical layer (PHY) data rate up to 600 Mbps. However, the 802.11 medium access control (MAC) layer control overhead, including the preamble, frame headers, carrier sense waiting time and the random backoff period, has limited the improvement of the user perceived throughput[2]. Especially with the high data
∗Correspondence to: Jui-Hung Yeh, Realtek-NCTU Joint Research Center, National Chiao Tung University, Hsinchu, Taiwan.
rate in 802.11n, MAC control overhead frequently occupies the channel when transmitting small-size frames, e.g. control messages and voice-over-IP (VoIP) traffic. In other words, the channel utilization is deteriorated. To enhance the channel utilization and network efficiency, 802.11n adopts the frame aggregation mechanisms that aggregate several frames into one single transmission.
Frame aggregation is particularly essential in the wireless mesh network (WMN) due to the repeated MAC control overhead in transmitting frames over the multi-hop wireless backhaul links[3]. A WMN is composed of mesh portals (MPPs), mesh points (MPs), mesh access points (MAPs) and wireless stations (STAs)[4]. MPPs, MAPs and MPs communicate with one another via the wireless medium and form a wireless backbone network. Moreover, STAs associate with an MAP to communicate with the other STAs or access the Internet via the MPP. Because of the nature of the wireless communication, only one mesh entity can transmit frames at a time within a collision domain[5, 6]. Besides, traffic loads of different mesh entities in the WMN are unbalanced[7]. The traffic load of the mesh nodes near the MPP is relatively high so that those nodes may have more frames to aggregate than the others. Appropriately aggregating frames can significantly moderate the MAC control overhead in the WMN.
In this paper, we propose a novel algorithm, namely dynamic aggregation selection and
scheduling (DASS), to maximize the bandwidth utilization and achieve a high-throughput and
high-efficiency WMN. It could dynamically select the aggregation mechanism according to the bit error rate (BER), the communication type, the transmission category, the quantity and the distribution of frame arrivals in the transmission queue. Based on the analysis of the traffic history, we could calculate the expected number of frame arrivals within a certain time. Then, an appro-priate waiting time is determined to aggregate incoming frames and send. We conduct extensive simulation to evaluate the throughput performance and the average delay of DASS. The results are to be compared with the fixed aggregation mechanisms.
The rest of this paper is organized as follows. Section 2 is dedicated to the related work. Section 3 provides an overview of the IEEE 802.11n frame aggregation mechanisms and 802.11s mesh networks. In Section 4, the DASS algorithm and its detailed operations are described. Section 5 presents the simulation results to evaluate the performance enhancement and provide proper configuration and parameters for DASS. Finally, Section 6 concludes this paper.
2. RELATED WORK
Wireless channel condition is commonly dynamic and error-prone. Transmission error can tremen-dously impact the performance of WLAN. Many literatures focus on modeling and analyzing the IEEE 802.11 CSMA/CA protocol. A two-dimensional Markov-chain is proposed to model the backoff behavior of saturated 802.11 WLAN by Bianchi [8]. Tay and Chua [9] propose a simple analytical model by using the average values for 802.11 WLAN. Yin et al.[10] analyze the throughput performance in error-prone channel and conclude that there could be an optimal frame size under a certain BER to achieve the maximum throughput. In [11] and [12], the analytical models are proposed to model frame aggregation of the CSMA/CA protocol. Furthermore, Kim
et al. [13] and Kuppa and Dattatreya [14] focus on modeling IEEE 802.11n frame
aggrega-tion mechanisms. Their results also confirm that frame aggregaaggrega-tion can enhance the throughput of WLAN.
Related work on adaptive frame aggregation schemes [15–18] aims to optimize the network performance by adjusting the aggregated frame size. The work in [15] leverages the 802.11e
priority queues to schedule the frame aggregation. In[16] an cross-layered adaptive frame aggre-gation scheme for WMN is proposed. Zhang et al. [18] propose a prediction-based link-layer dynamic frame reassembling scheme 802.11-based WMN. However, these two schemes above only aggregate MAC service data units (A-MSDUs) to enhance the scalability of WMNs. The 802.11n frame aggregation mechanisms are not considered either. Lin and Wong [17] conduct a thorough study of the newly proposed A-MSDU and A-MPDU frame aggregation schemes in IEEE 802.11n. They propose a simple optimal frame size adaptation algorithm for A-MSDU and A-MPDU under error-prone channels. Nevertheless, they only focus on enhancements of 802.11n MAC protocol, whereas the characteristics of mesh networks are not concerned. Therefore, the proposed algorithm could be inapplicable to WMN.
These existing studies do not consider how to choose an appropriate aggregation mechanism corresponding to the quantity and the distribution of frame arrivals, the communication types and transmission categories. Moreover, in WMN, the characteristics of the traffic vary with the roles and the locations of the communication pairs. To the best of our knowledge, there is no previous work collaborating the 802.11n frame aggregation mechanisms and 802.11s WMN. In this paper, we not only model the frame aggregation of CSMA/CA protocol in an error-prone channel to estimate the throughput of a given frame size but the proposed DASS can also dynamically predict and change the 802.11n frame aggregation mechanism according to the characteristics of the communication between the 802.11s mesh entities.
3. OVERVIEW OF FRAME AGGREGATION AND 802.11s WMNs
3.1. Frame aggregation mechanisms
Frame aggregation can be performed at different sub-layers [1]. Figure 1 illustrates the three major frame aggregation mechanisms—(1) A-MSDU, in which multiple MAC service data units (MSDUs) can be aggregated into a single MAC protocol data unit (MPDU) with a single MAC header; (2) aggregate MPDU(A-MPDU), which aggregates a number of MPDU to form a PHY service data unit (PSDU), (3) aggregate PHY protocol data unit(A-PPDU), which concatenates multiple PSDUs together and adds a PHY header.
Figure 2 illustrates the frame format of an A-MSDU that aggregates multiple MSDUs destined to the same destination and transmits them via a single MPDU. Because all MSDUs are concatenated into a single MPDU with a single frame control sequence (FCS), corruption of one subframe results in the failure of the entire A-MSDU. In 802.11n, A-MSDU increases the maximum frame transmission size from 2304 bytes to 7935 bytes. When there are many small MSDUs in the transmission queue, aggregating these small MSDU could significantly reduce the MAC control overhead. Thus, the channel efficiency could be improved.
The MPDUs that are physically transmitted to the same receiver could be aggregated into the A-MPDU. As shown in Figure 3, A-MPDU intends to concatenate the MPDUs to eliminate the PHY header followed by each MPDUs. The delimiter in each A-MPDU subframe assists the recovery of the A-MPDU when one or more received delimiters are with error. The value of the Duration/ID field in the MAC headers of all the MPDUs in the A-MPDU should be the same. In 802.11n, the maximum number of subframes it could hold is 64 and the maximum size is increased to 64k bytes.
Figure 4 depicts the format of an A-PPDU. A-PPDU concatenates multiple PSDUs with a common preamble to improve channel utilization. A-PPDU allows multiple frames to be sent to
MSDU
MPDU
PSDU
PPDU
MSDU SH
PHY SF MPDU SF MPDU ... SF MPDU MAC LLC MPDU Aggregation MSDU Aggregation PHY PPDU Aggregation
PHY MAC SH MSDU ... FCS
PHY DL MPDU DL MPDU ... DL MPDU
DL: Delimiter
SF: Signal field
MPDU
Figure 1. Layers of WLAN interface.
... Subframe N Subframe 2 Subframe 1 SA Length Padding A D MSDU PSDU Subframe HDR
Figure 2. The frame format of an A-MSDU.
different destination addresses in one single medium access. Different PSDUs within the A-PPDU are separated by a Physical Layer Convergence Procedure (PLCP) signal field.
The comparison among the three frame aggregation mechanisms is summarized in Table I[19]. A-MSDU has the characteristics of the highest network efficiency because there is no additional MAC control overhead for each MSDU. However, MSDUs are restricted to a single pair of source and destination endpoints and traffic class. A-MSDU is also vulnerable to transmission errors owing to that the payload of an A-MSDU is merely protected by a single cyclic redundancy check. A-MPDU and A-PPDU have the advantage of aggregating frames destined to multiple destination addresses. Moreover, each frame within the A-MPDU or A-PPDU is individually attached with an FCS that can ensure the correctness of frame transmission. Thus, A-MPDU and A-PPDU are relatively robust to the transmission error. A-PPDU has another advantage that the PPDUs destined to different destination addresses may adopt different transmission rates with different modulations.
... Subframe N Subframe 2 Subframe 1 Delimiter Signature MPDU Length CRC Reserved PSDU U D S M A R D H Y H P MPDU Padding MPDU Delimiter 4 Bytes:
Figure 3. The frame format of an A-MPDU.
PHY SYNC
HDR SIGNAL MPDU MPDU MPDU
RIFS SIGNAL RIFS SIGNAL
Figure 4. The frame format of an A-PPDU.
Table I. Comparisons of frame aggregation mechanisms[19].
A-MSDU A-MPDU A-PPDU
MAC control overhead Low Middle High
Network efficiency enhancement High Middle Low
Multiple destination addresses No Yes Yes
Robustness for transmission error Low High High
Rate change for different destination No No Yes
3.2. IEEE 802.11s mesh networks
IEEE 802.11s defines the mesh networking using the IEEE 802.11 MAC/PHY layers that support layer-2 path selection protocols and data forwarding over multi-hop topologies. Figure 5 illustrates the architecture of a mesh network.
Based on the roles of the mesh entities and the transmission behaviors, we first consider three types of communication pairs and four categories of transmission in the WMN. The peer-to-peer communication among the wireless mesh nodes could be classified into three types, including M(A)P-to-M(A)P, MAP-to-STA and STA-to-MAP. In addition, in WMN, the receiver of a peer link might not be the destination of the frame. Thus, frames to different destinations could be aggregated into one single frame. There are four transmission categories for an aggregated frame in this multi-hop environment, namely single destination single receiver (SDSR), multiple destination single receiver (MDSR), multiple destination multiple receiver (MDMR) and single destination multiple receiver (SDMR).‡ According to the characteristics of the communication pair and the
Figure 5. IEEE 802.11s mesh networks architecture.
Table II. The adaptive frame aggregation mechanisms among different transmission pairs.
MP-to-MP STA-to-MAP MAP-to-STA
SDSR MDSR SDMR MDMR SDSR MDSR SDSR MDMR
A-MSDU
A-MPDU
A-PPDU
: Most suitable, : Replaceable but less efficient, : Replaceable but inefficient.
transmission category, the corresponding suitable aggregation mechanisms can be determined. For example, A-PPDU is suitable for the MDMR case because it is the only applicable aggregation mechanism for multiple destinations and receivers. A-PPDU is not suggested for the STA-to-MAP case because the MAP is the only receiver in this case. But multiple STAs may associate with one MAP at the same time. Therefore, A-PPDU could be adopted by an MAP to aggregate the frames destined to different STAs, i.e. MAP-to-STA.
The usage of a frame aggregation mechanism varies with the roles of the communication pair. Table II shows the usage of the aggregation mechanisms in different communication scenarios. A-MSDU reduces the most MAC control overhead. In the aspect of the network efficiency enhance-ment, A-MSDU is the most suitable aggregation mechanism for all the SDSR cases. However, MSDU is restricted to the SDSR case. Also, if the BER is high, the user could choose A-MPDU rather than A-MSDU to prevent the entire A-MSDU from corruption. For the MDSR cases, A-MPDU could improve more network efficiency than A-PPDU because of the saved control overhead. Although A-PPDU eliminates the least MAC and PHY overhead and is not suggested if A-MSDU and A-MPDU are applicable, it is the only choice for the cases of multiple receivers, i.e. SDMR and MDMR.
Figure 6. The flow chart of the DASS algorithm.
4. DASS ALGORITHM
This section details the concepts and procedures of the proposed DASS algorithm. The DASS algorithm decides which aggregation mechanism to employ and when to transmit the frames. The expected quantity and distribution of the frame arrivals in the transmission queue are evaluated. Based on the prediction of the frame arrivals, DASS dynamically selects the preferable frame aggregation mechanism and schedules the transmission to achieve a throughput and high-efficiency mesh network.
4.1. Overview of the algorithms
There are four phases in the proposed DASS algorithm. In the first phase of DASS, each aggregation point filters out the inappropriate aggregation mechanisms before transmission. In the latter phases of DASS, the channel quality, the quantity, distribution and the expected arrival rate of the incoming frames are measured and evaluated. Based on these factors, DASS algorithm determines: (1) which aggregation mechanism to adopt and (2) when to send the aggregated frame out. The operations of the algorithm are depicted in Figure 6 and elaborated in the following subsections.
4.2. Detailed operations of DASS
4.2.1. First phase: filtering out inappropriate aggregation mechanisms. When a mesh node joins a
DASS can filter out the inappropriate aggregation mechanisms in the first phase before each transmission. In this paper, every STA is supposed to follow the 802.11 standard to have only one link to its associated MAP in the Basic Service Set (BSS). Thus, if a transmitter is an STA, DASS will not consider A-PPDU to aggregate the frames because multiple receivers, MDMR, will not happen to such a transmission.
4.2.2. Second phase: getting the optimal frame size. In the second phase, DASS analyzes the
optimal frame size for the available aggregation mechanisms. We adopt Lin and Wong’s analytical model[17] that extends the model from Bianchi [8] to compute the optimal frame size under error-prone channels. In their analytical model, N mobile stations are assumed to exist in the WLAN. In mesh networks, the BSS traffic and the mesh forwarded traffic compete for the same channel that can only be occupied by one channel access at a time. Thus, N is redefined as the number of all mesh nodes which can sense each other within the same collision domain. The wireless channel is assumed to be with a bit-error-rate (BER) of Pb, which can be measured through an
incoming frame. The minimum contention window size is W and the maximum backoff stage is m. Because the size of an aggregated frame is large, the request-to-send/clear-to-send (RTS/CTS) access scheme is enabled all the time. Moreover, the control frames are transmitted at the basic rate, which is much more robust in combating errors than the data frames transmitted at the data rate. To simplify the analysis, the analytical model does not consider the frame error for control frames and preambles.
In[8, 17], a virtual time slot system is adopted to model the state of the queuing system. Each virtual time slot represents the interval between two consecutive countdown of backoff timers by non-transmitting stations.
From[17], the transmission probability in a virtual time slot is
= 2(1−2p)
(1−2p)(W +1)+ pW(1−(2p)m) (1)
where p is the conditional probability that an unsuccessful transmission happens in a time slot. p can further be expressed as
p=1−(1−(1−(1−)(N−1)))(1− pe) (2)
where pedenotes the conditional error probability that an error happens after a successful RTS/CTS
exchange in the time slot.
The probability of an idle slot is
Pidle=(1−)N (3)
The transmission probability for a time slot is
Ptr=1− Pidle=1−(1−)N (4)
Thus, the conditional probability for a non-collided transmission is
Ps=
N(1−)(N −1)
Ptr
(5) The probability that the transmission fails not due to collisions but transmission errors is
Similarly, the probability for a successful transmission without collisions and transmission errors is
Psucc= PtrPs(1− pe) (7)
The saturated network throughput can be derived as
S=Ep Et
(8) where Ep is the number of bits of payload that are successfully transmitted in a virtual time slot. Et denotes the expected time length of a virtual time slot. Et can be expressed as
Et= TidlePidle+TcPtr(1− Ps)+Teperr+TsuccPsucc (9)
where Tidle, Tc and Tsuccare the lengths of the idle, collision and successful virtual time slot. Teis
the virtual time slot length for an error transmission sequence. Tidleis the default empty slot time
of the system.
The average delay of channel access is then derived as
d= NLp
S (10)
where Lpis payload length of the aggregated frame.
In this paper, we only consider the uni-directional case in [17]; the equations for Tc, Tsucc, Te
and Epare
Tc= RTS+EIFS (11)
Tsucc= RTS+CTS+DATA+BACK+3SIFS+DIFS (12)
Te= RTS+CTS+DATA+EIFS+2SIFS (13)
For A-MSDU, pe and Epcan be obtained by
pe= 1−(1− Pb)L (14)
Ep= (L − Lhdr)(1− pe) (15)
where L is the aggregated MAC frame size and Lhdr is the total length including MAC header
and FCS.
For A-MPDU, error occurs when all the sub-frames become corrupted. Thus, we have
pe= i (1−(1− Pb)Li) (16) Ep= i (Li− Lsubhdr)(1− Pb)Li (17)
Table III. Notations of the third phase of DASS.
Endpointi, j The set of frames which are destined to the endpoint i with the TID equals to j , where 0j7
f(x,BER) The goodput function that is defined by divide the optimal aggregated frame size
f(x,BER) x to the duration of sender successfully transmitting the frame under the BER Dbuffered(i, j) The amount of the buffered data for Endpointi, j
Rrm The set of the traffic that will be transmitted to the same receiver m DRr(m) The amount of the buffered data for Rrm
TMax−MSDU The current maximum throughput using A-MSDU TMax−MPDU The current maximum throughput using A-MPDU TMax−PPDU The current maximum throughput using A-PPDU
where i is from 1 to the total number of aggregated sub-MPDUs and Li is the size for the i th
sub-MPDU. Lsubhdr is the total size of each sub-MPDU’s delimiter, header and FCS.
To summarize, according to the given aggregation method and the frame size, DASS can evaluate the expected network throughput S and the average access delay d from Equation (8) and (10). According to the value of S, the optimal aggregated frame size can be derived.
4.2.3. Third phase: performance evaluation. After getting the optimal frame size of available
aggregation mechanisms, the third phase of DASS intends to select an aggregation mechanism with the largest throughput improvement for the mesh node.
In the second phase, one can observe that the optimal aggregated frame size varies with BER. Since a mesh node may have more than one neighbor, it is necessary to consider the diverse BER between different communication peers. To explain the third phase of DASS, we take an example that a mesh has a certain number of frames destined to the endpoint i with TID equals j . Table III defines the notations used in this phase.
When frames are enqueued in the transmission queue, DASS, based on the available aggregation mechanisms, shall, respectively, compute the maximum throughput. All of the frames in the transmission queue are classified according to destination address and TID value.
For A-MSDU, frames could only be aggregated when their destination and TID value are the same. BER measured between the communication pair along with the accumulative size of the frames, x, could then be used as the input of f(x,BER) to calculate the corresponding throughput. We repeat this procedure for each set of aggregated frames, Endpointi, j. The maximum throughput of the transmission queue under A-MSDU is then obtained. Note that different set of aggregated frames may have the same maximum throughput. The expected maximum throughput can be formu-lated as
TMax−MSDU=max
i, j { f (Dbuffered(i, j),BER)} (18)
For A-MPDU, the frames with the same receivers can be aggregated. Via routing information, each mesh node can classify all the frames in transmission queue according to the next-hop receivers. In A-MPDU, frames can be aggregated as long as having the same receiver. Each mesh node might determine the concatenatability of individual nodes according to the routing information. In a similar way, the maximum throughput of the transmission queue can be obtained. Note that as in A-MSDU, different set of concatenatable frames may have the same maximum
throughput as
TMax−MPDU=max
m { f (DRr(m),BER)} (19)
A-PPDU has no restrictions on concatenation. The maximum throughput is computed in a similar way, except that it is the maximum among all possible frame aggregations as
TMax−PPDU= f i 7 j=0
DBuffered(i, j),BER
(20)
Through the comparison among the three maximum throughput values TMax−MSDU, TMax−MPDU
and TMax−PPDU, DASS can determine the aggregation mechanism and proceed to the next phase.
4.2.4. Fourth phase: scheduling packets. In the second stage, different BER values bring different
optimal aggregated frame sizes of the corresponding aggregation mechanisms. The optimal aggre-gated frame size is called the ideal value. Afterward, through the third stage, we can estimate the maximal throughput of the different aggregation mechanisms and determine which aggregation mechanism to adopt. Comparing the ideal values with the accumulative amount of frames, there are three possible situations.
The first situation is that the current amount of frames is greater than the ideal value. If the amount of frames is greater than the ideal value, the frames will be aggregated to a size close but smaller than the ideal value. For A-MSDU, the selection strategy is first-in-first-out (FIFO). On the other hand, for A-MPDU and A-PPDU, the selection depends on the QoS classes. The frames with the higher QoS type have the higher priority to be sent. If the frames are with the same QoS type, DASS selects the frames that have traversed more hop-counts before arriving the current aggregation point. This selection strategy aims to minimize the variation of the end-to-end latency among all frames.
The second situation is that the amount of frames is equal to the ideal value. When the amount of frames is equal to ideal value, the only choice is to aggregate these frames and then transmit it.
The third situation, which is the most complicated, is that the amount of frames is less than the ideal value. When the amount of frames is less than the ideal value, DASS, based on the past traffic, predicts the arrival rate of the same set of frames. According to the past 16 frame arrivals, DASS could estimate the frame arrival rate by dividing the accumulative amount of frames to the time duration of the past 16 frames. The expected arrival rate, RPredict, is formulated as
RPredict(i, j)=
16Lp
16
k=1Ak
(21)
where Lp is the payload length of the aggregated frame and Ak is the inter-arrival time between
framek and framek+1.
By computing the frame arrival rate, we could estimate whether the aggregation of the frames of Endpointi, j could improve the throughput. The expected throughput of transmitting the aggregated buffered data, ThBuffered, is
Assuming that TWaiting is the waiting time for the incoming frames, the total incoming frame
sizes could be calculated by the frame arrival rate as
DFuture(i, j)= RPredict(i, j)·TWaiting (23)
Then we could calculate the predicted throughput of transmitting the aggregated frame after waiting TWaiting seconds. The equations of the expected throughput of the three aggregation
methods, namely ThPredict A−MSDU, ThPredict A−MPDUand ThPredict A−PPDUare derived as ThPredict A−MSDU= min(DBuffered(i, j)+ DFuture(i, j),7935)
DBuffered(i, j)+ DFuture(i, j)
f(DBuffered(i, j)+ DFuture(i, j),BER)+TWaiting
(24)
ThPredict A-MPDU=
min(7j=0(DBuffered(i, j)+ DFuture(i, j)),64k)
7
j=0(DBuffered(i, j)+ DFuture(i, j)) f(7j=0(DBuffered(i, j)+ DFuture(i, j)),BER)
+TWaiting (25) ThPredict A-MPDU= i 7
j=0(DBuffered(i, j)+ DFuture(i, j))
i
7
j=0(DBuffered(i, j)+ DFuture(i, j)) f(i7j=0(DBuffered(i, j)+ DFuture(i, j)),BER)
+TWaiting
(26)
Afterward, DASS compares the ThBuffered with ThPredict∗ (ThPredict∗∈ {ThPredict A−MSDU,
ThPredict A−MPDU, ThPredict A−MPDU}) to decide whether the MAC should wait for the more frames for aggregation or not. If ThPredict∗>ThBuffered, DASS triggers an internal timer with timer length
equal to TWaiting. Then, the MAC suspends the transmission and waits for the frame arrivals before
the timer runs out. If the amount of the follow-up frames is larger than the maximum size of the aggregated frame, DASS stops the timer and orders the MAC to transmit the frames with the corresponding frame aggregation mechanism. If ThPredict∗ThBuffered, the MAC will immediately
aggregate the frames in the transmission queue and send out.
5. SIMULATION RESULTS
This section verifies the performance of DASS through the extensive simulations with the ns-2[20] simulator. Section 5.1 first describes the simulation environment. Moreover, Section 5.2 elaborates the simulation results. The throughput performance under infinite and steady backlog, the accuracy of prediction for frame arrival rate and the comparisons between different selection strategies are presented.
5.1. Simulation environment
In the simulation, we set up an infrastructure service area that includes 8 MPs, 8 MAPs and 10–30 STAs. To simulate the IEEE 802.11n MAC, we set the basic rate to 54 Mbps to transmit the management frames, whereas the data frame are transmitted with 144.44 Mbps[1]. Based on the 802.11n, all entities operate over a 20 MHz channel with the same modulation coding scheme.
Table IV. Simulation parameters. Parameter Value Basic rate 54 Mbps Data rate 144.44 Mbps PLCP preamble 16s PLCP header 48 bits PLCP rate 6 Mbps
MAC header 192 bits
FCS (Frame check sequence) 32 bits
Time slot 9s
Sub-frame header in A-MSDU 14 Bytes
Delimiter in A-MPDU 4 Bytes
Duration of signal field in A-PPDU 4s
RIFS (Reduced inter frame space) 2s
SIFS (Short inter frame space) 16s
DIFS (Data inter frame space) 34s
Size of ACK frame 14 Bytes
Size of block ACK frame 32 Bytes
These devices are placed over a distance of 50 m with line of sight (LOS) communications. Each STA generates traffic with a predefined rate. These sources have no timeout values specified and they might have different TID. To simulate the real-time traffic, all the data packets pass down to the MAC layer are 100 bytes in length. The BER varies from 0 to 10−3. The simulation time is 10 s. The detailed MAC parameters used in the simulation are shown in Table IV.
5.2. Simulation results
5.2.1. Throughput enhancement. First, we discuss the throughput enhancement by adopting the
DASS algorithm. The simulation is carried out with the saturated traffic of 10–30 STAs. We compare the throughput improvement of DASS with those of the hybrid without waiting and single frame aggregation schemes. For the purpose of comparison, a simple hybrid frame aggregation scheme that transmits frames immediately after selecting a suitable frame aggregation mechanism is elaborated. Figure 7 shows the throughput improvement by adopting various frame aggregation schemes under the saturated traffic. The results showed that the hybrid aggregation sc heme apparently outperforms the other single aggregation schemes. Moreover, DASS still has better throughput improvement than the simple hybrid frame aggregation scheme. This could be because an STA might send an A-MSDU aggregated frame to its associated MAP, but, in DASS, the associated MAP might wait for more frames to use A-MPDU or A-PPDU for a larger aggregated frame and better throughput. Comparing with the case without any frame aggregation, DASS could enhance the overall throughput by 92%. Another phenomenon we have observed is that the degree of throughput improvement decreases slightly with an increase in the number of STAs. The reason is that when the channel is completely saturated, the time wasted on a CSMA/CA random backoff and the probability of collisions might be raised with an increase in contention for bandwidth. The contention and retransmission then deteriorate the throughput.
Next, we discuss the importance of the waiting mechanism in DASS, especially when the traffic load is not saturated. We conduct the simulation with the unsaturated traffic and 10–30 STAs. Figure 8 presents the throughput improvement for the various frame aggregation schemes.
Figure 7. Throughput improvement by adopting frame aggregation schemes under the saturated traffic.
The simulation results show that the degree of throughput improvement of DASS, which enables the waiting strategy, apparently outperforms those schemes without the waiting mechanism. Comparing with the case with no frame aggregation, DASS could enhance the overall throughput by 95%. However, please note that, different from the former experiment, the channel is not saturated in this experiment. The degree of throughput improvement increases as the number of STAs increases. Therefore, channel utilization is increased when the traffic load was getting saturated.
5.2.2. Accuracy of prediction of frame arrival rate. Through the prediction of frame arrivals,
DASS can analyze and decide whether to wait for the follow-up frames to aggregate to seek better throughput. The results in Section 5.2.1 demonstrate that adopting the waiting mechanism can further enhance the degree of throughput improvement. However, how many previous frames should the DASS algorithm consider for the prediction of the frame arrivals? In this section, an experiment is designed to observe the success rate, which is defined as
Success rate=Counts of improved throughput
Counts of waiting (27)
Table V lists the success rate of the waiting strategy versus the number of past frames observed. Because the traffic of STAs is more stable than the other entities, the success rate in the STAs is relatively high and approaches to 92.53%. Moreover, the success rate of each aggregation point commonly decreases when the number of the referred past frames increases. The degree of degradation is especially severe and evident for the STAs. This phenomenon is caused by the
On–Off behavior of the wireless traffic flow. If the frames generated from the STAs are transmitted
Figure 8. Throughput improvement by adopting frame aggregation schemes under the unsaturated traffic.
Table V. Success rate of the waiting strategy of the mesh entities.
STA MAP MP
Generating traffic Yes No No
Traffic load Low Middle High
Success rate—past 16 frames (%) 92.53 82.74 89.6
Success rate—past 32 frames (%) 93.25 83.42 89.87
Success rate—past 64 frames (%) 86.8 81.78 88.72
Success rate—past 128 frames (%) 82.73 79.35 86.64
should converge to a fixed value gradually with the increase in the number of the referred past frames. Figure 9 illustrates the average throughput of DASS when referring to different numbers of the previous frames. Based on the success rate observed in Table V, the throughput of referring to the past 16 and 32 frames is relatively better when the prediction of frame arrival rate is more precise. In other words, the wasted waiting time caused by the unsuccessful prediction degrades the throughput.
5.2.3. Comparisons between different selection strategies. In this section, we discuss the selection
strategies that come up when there are sufficient frames from different sets of traffic, Endpointi, j, to be aggregated at the same time. The experiment compares five selection strategies in terms of average latency and throughput. Specifically, the hop-count-based strategy depends on the hop count traversed from the source to the current aggregation point. A frame with more hop count has a higher priority for aggregation. The propagation-delay-based strategy first aggregates the frames
Figure 9. Throughput comparison of referring different numbers of the past frames.
Figure 10. Comparison on the average latency among the different selection strategies.
with more propagation delay from the source to the current aggregation point. Moreover, for the First Come First Served (FCFS) strategy, DASS simply aggregates the frames in the manner of FCFS. Also, the ratio-based strategy selects the frames from the set of traffic that has the largest amount. Finally, the channel-quality-based strategy selects the frames that possess the best channel quality between the receiver and the current aggregation point. Please note that the strategies for different purposes can be mixed to seek for better performance. For example, we could combine the propagation-delay-based and the channel-quality-based strategies.
Figures 10 and 11 present the comparisons of the average latency and the throughput performance for the five selection strategies. In Figure 10, based on the channel quality between the aggregation point and the receiver to select the aggregated queue would decrease the average latency most. Therefore, the throughput performance is also largely improved as shown in Figure 11. Note that the channel quality here is exactly the BER value measured in the second stage of DASS. Besides, we observe that the variance of the average latency for different frame selection strategies is not
Figure 11. Comparison on the average throughput among the different selection strategies.
significant. Analyzing the variation of the latency, we could find that the variance of latency for the FCFS strategy is the highest, whereas the propagation-delay-based strategy is the lowest.
6. CONCLUSIONS
Frame aggregation mechanisms are introduced to resolve the fundamental problem of existing IEEE 802.11 MAC overhead. This paper proposes the DASS algorithm that dynamically adopts an appropriate aggregation mechanism according to the characteristics of communications to maximize the bandwidth utilization. It also eases the throughput degradation drawn by multi-hop transmissions in the WMNs. Simulation results demonstrate that DASS could improve the overall throughput of 802.11s mesh network by 95% compared with the case with no aggregation. It is also suggested that DASS should refer the past 16 or 32 frames for the prediction of the frame arrivals. Moreover, the channel-quality-based selection strategy could reduce latency most and significantly enhance the throughput. Overall, DASS is an innovative algorithm that collaborates the 802.11n frame aggregation mechanisms to achieve a high-throughput and high-efficiency WMN.
REFERENCES
1. IEEE P802.11n/D2.0. Amendment: medium access control (MAC) and physical layer (PHY) specifications, enhancement for higher throughput. IEEE P802.11n/D2.0, March 2007.
2. IEEE Std 802.11. Part 11: wireless LAN medium access control (MAC) and physical layer (PHY) specifications. IEEE Std 802.11, June 2007.
3. Akyildiz IF, Wang X, Wang W. Wireless mesh networks: a survey. Computer Networks 2005; 47(4):445–487. 4. IEEE P802.11s/D1.06. Draft amendment to standard IEEE 802.11: mesh networking. IEEE P802.11s/D1.06,
May 2007.
5. Jun J, Sichitiu ML. The nominal capacity of wireless mesh networks. IEEE Wireless Communications 2003;
10:8–14.
6. Jain K, Padhye J, Padmanabhan V, Qiu L. Impact of interference on multi-hop wireless network performance. Proceedings of ACM/IEEE International Conference on Mobile Computing and Networking (MobiCom), San Diego, CA, U.S.A., September 2003.
7. Bicket J, Aguayo D, Biswas S, Morris R. Architecture and evaluation of an unplanned 802.11b mesh network. Proceedings of ACM/IEEE International Conference on Mobile Computing and Networking (MobiCom), Cologne, Germany, August 2005.
8. Bianchi G. Performance analysis of the IEEE 802.11 distributed coordination function. IEEE Journal on Selected Areas in Communications 2000; 18:535–547.
9. Tay YC, Chua K. A capacity analysis for the IEEE 802.11 MAC protocol. ACM Wireless Networks 2001;
7(2):159–171.
10. Yin J, Wang X, Agrawal DP. Optimal packet size in error-prone channel for IEEE 802.11 distributed coordination function. Proceedings of IEEE Wireless Communications and Networking Conference, Atlanta, GA, U.S.A., March 2004.
11. Lu K, Wu D, Qian Y, Fang Y, Qiu RC. Performance of an aggregation-based MAC protocol for high-data-rate ultrawideband ad hoc networks. IEEE Transactions on Vehicular Technology 2007; 56:312–321.
12. Lu K, Wang J, Wu D, Fang Y. Performance of a burst-frame-based CSMA/CA protocol: analysis and enhancement. ACM Wireless Networks 2009; 15:87–98.
13. Kim Y, Choi S, Jang K, Hwang H. Throughput enhancement of IEEE 802.11 WLAN via frame aggregation. Proceedings of IEEE Vehicular Technology Conference, Los Angeles, CA, U.S.A., September 2004.
14. Kuppa S, Dattatreya G. Modeling and analysis of frame aggregation in unsaturated WLANs with finite buffer stations. Proceedings of IEEE International Conference on Communications(ICC), Istanbul, Turkey, June 2006. 15. Lin Y-S, Wang J-Y, Hwang W-S. Scheduling mechanism for WLAN frame aggregation with priority support.
Proceedings of Joint Conference on Information Sciences(JCIS), Kaohsiung, Taiwan, October 2006.
16. Riggio R, Pellegrini FD, Scalabrino N, Li P, Fang Y, Chlamtac I. Performance of a novel adaptive traffic aggregation scheme for wireless mesh networks. Proceedings of IEEE Military Communications Conference (MILCOM 2007), Orlando, FL, U.S.A., October 2007.
17. Lin Y, Wong VWS. Frame aggregation and optimal frame size adaptation for IEEE 802.11n WLANs. Proceedings of IEEE Global Telecommunications Conference(GLOBECOM), San Fransisco, CA, U.S.A., November 2006. 18. Zhang L, Dong W, Wang G. Using dynamic frame reassembling to improve transmission performance in wireless
mesh networks. International Journal of Intelligent Information Technology Application 2008; 1:86–91. 19. Nagai Y, Fujimura A, Shirokura Y, Isota Y, Ishizu F, Nakase H, Kameda S, Oguma H, Tsubouchi K. 324 Mbps
WLAN equipment with MAC frame aggregation for high MAC-SAP throughput. Journal of Communications 2005; 1:1–8.
20. The network simulator—ns-2. Available from: http://www.isi.edu/nsnam/ns/.
AUTHORS’ BIOGRAPHIES
Ying-Dar Lin received the Bachelor’s degree in Computer Science and Information
Engineering from the National Taiwan University in 1988, and the MS and PhD degrees in Computer Science from the University of California, Los Angeles (UCLA) in 1990 and 1993, respectively. He joined the faculty of the Department of Computer Science at the National Chiao Tung University (NCTU) in August 1993 and is Professor since 1999. He spent his sabbatical year as a visiting scholar at Cisco, San Jose, in 2007–2008. He was the director of Institute of Network Engineering during 2005–2007. He is also the founder and director of Network Benchmarking Lab (NBL), co-hosted by Industrial Technology Research Institute (ITRI) and NCTU since 2002, which reviews the functionality, performance, conformance and interoperability of networking products ranging from switch, router, WLAN to security and VoIP. In 2002, he co-founded L7 Networks Inc., which addresses the content networking markets with the technologies of deep packet inspection. At NCTU, he currently directs, or co-directs, Computer and Network Center (2007), NBL (2002), Realtek-NCTU Joint Lab (2006) and D-Link NCTU Joint Lab (2007). His research interests include design, analysis, implementation and benchmarking of network protocols and algorithms, quality of services, network security, deep packet inspection, embedded hardware software co-design, WLAN mesh and P2P networking. From 2008, he is on the editorial board of IEEE Communications Magazine and IEEE Communications Surveys and Tutorials. He was on the program committee of ICCCN’07 and a program co-chair of International Computer Symposium ’07. He has been a consultant for several high-tech companies
and authored a textbook on Computer Network Experiments. Dr Lin is a senior member of IEEE. He can be reached at ydlin@cs.nctu.edu.tw.
Jui-Hung Yeh received the PhD degree from the Department of Computer Science,
National Tsing Hua University (NTHU), Hsinchu, Taiwan, in 2008. He is currently a postdoctoral researcher with the Realtek-NCTU Joint Research Center, National Chiao Tung University (NCTU), Hsinchu, Taiwan. His research interests include mesh network, QoS, mobility management, integration of wireless networks and performance evaluation. He is a member of IEEE.
Tsung-Hsien Yang received his MS degree from the Department of Computer Science,
National Chiao Tung University, Hsinchu, Taiwan, in 2008. He is now in his military service until 2009.
Chia-Yu Ku received his MS degree from the Department of Computer Science, National
Chiao Tung University, Hsinchu, Taiwan, in 2008. He is earning his PhD degree since September 2007. His research interests include wireless mesh networking, multi-channel multi-radio, mobility management, integration of wireless networks and performance evaluation.
Shiao-Li Tsao received BS, MS, and PhD degrees in engineering science from the
National Cheng Kung University, Taiwan, in 1995, 1996 and 1999, respectively. His research interests include mobile communication and wireless network, and embedded software and system. He is currently an assistant professor of computer science of National Chiao Tung University. Prof. Tsao has published more than 60 international journals and conference papers, and received the Outstanding Project Award of Ministry of Economic Affairs (MOEA) of Taiwan, in 2003, the Advanced Technologies Award of MOEA in 2003, Young Researchers Award of Pan Wen-Yuan Foundation in 2007 and Young Engineer Award of the Chinese Institute of Electrical Engineering in 2008. He is a member of IEEE.
Yuan-Cheng Lai received his PhD degree in the Department of Computer and
Infor-mation Science from the National Chiao Tung University in 1997. He joined the faculty of the Department of Computer Science and Information Science at the National Cheng Kung University in August 1998. He then joined the faculty of the Department of Information Management at the National Taiwan University of Science and Technology in August 2001 and has been a professor since February 2008. His research inter-ests include performance analysis, protocol design, wireless networks and Web-based applications.
![Table I. Comparisons of frame aggregation mechanisms [19].](https://thumb-ap.123doks.com/thumbv2/9libinfo/7603752.129276/5.850.120.735.521.635/table-i-comparisons-of-frame-aggregation-mechanisms.webp)
