[ieee 2011 international conference on control, automation and systems engineering (case) -...

5
A Distributed Cross-layer Adaptive Scheduling Algorithm for Wireless Mesh Networks Hua GUO Dept. of Communication Engineering Tianjin Polytechnic University Tianjin, China [email protected] Jian GUO School of Management Tianjin University of Science Tianjin, China Zhong ZHANG Div. of Science & Technology Tianjin University Tianjin, China Jiansong ZHANG Postbox 2861-6 Beijing, China Abstract—In this paper, we propose a cross-layer scheduling scheme named DCLASA for the coordinated distributed scheduling of IEEE 802.16 WMN. This scheme considers the channel quality information (CQI) in physical layer when designing the timeslot allocation algorithm in MAC layer, and adjusts the node’s holdoff exponent according to CQI. Meanwhile multi-user diversity and adaptive modulation and coding technique are adopted to improve system performance. Simulation shows that DCLASA can increase the network total throughput, and reduce link transmission delay. Keywords- IEEE 802.16 Mesh networks; cross-layer design; coordinated distributed scheduling; DCLASA I. INTRODUCTION Wireless Mesh Network (WMN) is a cost-effective solution for building broadband Internet access and backhaul systems. The popularity of WMNs is due to the flexible network architecture, easy deployment, and self- healing capability [1]. Several medium-access control (MAC) schemes have been developed for WMNs, namely contention-based, schedule-base, and hybrid MAC. The carrier-sense multiple access (CSMA) protocol used by IEEE 802.11WMNs is contention-based. It is prevailing in early periods for its simplicity and good throughput. Unfortunately, hidden terminal problems degrade its performance of substantially, because collisions can’t be prevented. Schedule-based transmission scheme works in a way that no collisions occur, for example, TDMA and FDMA. In this approach, resources such as time slots, frequencies, and spreading codes are explicitly allocated to network stations. Normally, in schedule-based MAC, a centralized base station is required to make resource assignment for subscriber stations. However, over the past decade, a number of protocols have been proposed to provide distributed time-slot allocation. In REALM [2], once all the nodes in the two-hop neighborhood of a node have consistent information regarding their neighborhood, collisions are avoided because all the nodes use the same deterministic algorithm to select a winner for a given time slot. A more efficient approach than REALM is to let those nodes that were able to transmit during a prior frame notify their neighborhoods that they will not be competing for transmission slots for a period of time, which effectively reduces contention among nodes and the delay experienced by a given node in accessing the channel [3]. For example, the Coordinated Distributed Scheduling (CDS) of IEEE 802.16 mesh mode introduces holdoff mechanism to reduce contention. In this paper, we focus on CDS of 802.16 mesh mode for it is a comprehensive expression of up-to-date schedule-based WMN research. Former researches on 802.16 CDS are mainly about adjusting holdoff time by considering node’s data queue length [4], number of neighbors[5], transmission state (sending, receiving, forwarding, or idle) [6], and whether to have three-way handshake information to send [7]. However, the SNR of different wireless channel is not taken into account, which will lead to inefficiency in time varying communication environment. Recently paper [8] begins to pay attention to that by adjusting modulation scheme to improve the band utilization for the mesh SSs with fine channel quality, but it doesn’t utilize dynamic holdoff time to control mesh contention. Moreover, WMNs frequently act as the backhaul solution for a large number of access points serving a given coverage area. Therefore, the traffic loads among nodes of WMN is quite high. However, previous researches don’t pay much attention to this, and normally it is assumed that only partial links of WMNs are busy. For example, the simulation scenario is being devised that there are only 2 to 5 concurrent flows in the network, or the sending queue of a few busy nodes are not empty. When the backhaul WMN are fully occupied, the assumption of former research might be false. This results in failure of optimization. To overcome the drawbacks of existing research mentioned above, in this paper, a distributed cross-layer adaptive scheduling algorithm (DCLASA) for WMN is proposed. The outline of this paper is as follows. Section II presents a survey of existing research on the coordinated distributed scheduling of WMN. Section III gives the system model and the algorithm description. Section IV shows the simulation results and a detailed analysis. Section V concludes the paper. II. COORDINATED DISTRIBUTED SCHEDULING OF WMN A. Mathimatical model of 802.16 mesh coordinated distributed scheduling In this section, we briefly review the 802.16 mesh CDS from the three-way handshake perspective. The bandwidth 978-1-4577-0860-2/11/$26.00 ©2011 IEEE

Upload: jiansong

Post on 13-Dec-2016

213 views

Category:

Documents


1 download

TRANSCRIPT

A Distributed Cross-layer Adaptive Scheduling Algorithm for Wireless Mesh Networks

Hua GUO Dept. of Communication Engineering

Tianjin Polytechnic University Tianjin, China

[email protected]

Jian GUO School of Management

Tianjin University of Science Tianjin, China

Zhong ZHANG

Div. of Science & Technology Tianjin University

Tianjin, China

Jiansong ZHANG Postbox 2861-6 Beijing, China

Abstract—In this paper, we propose a cross-layer scheduling scheme named DCLASA for the coordinated distributed scheduling of IEEE 802.16 WMN. This scheme considers the channel quality information (CQI) in physical layer when designing the timeslot allocation algorithm in MAC layer, and adjusts the node’s holdoff exponent according to CQI. Meanwhile multi-user diversity and adaptive modulation and coding technique are adopted to improve system performance. Simulation shows that DCLASA can increase the network total throughput, and reduce link transmission delay.

Keywords- IEEE 802.16 Mesh networks; cross-layer design; coordinated distributed scheduling; DCLASA

I. INTRODUCTION Wireless Mesh Network (WMN) is a cost-effective

solution for building broadband Internet access and backhaul systems. The popularity of WMNs is due to the flexible network architecture, easy deployment, and self-healing capability [1].

Several medium-access control (MAC) schemes have been developed for WMNs, namely contention-based, schedule-base, and hybrid MAC. The carrier-sense multiple access (CSMA) protocol used by IEEE 802.11WMNs is contention-based. It is prevailing in early periods for its simplicity and good throughput. Unfortunately, hidden terminal problems degrade its performance of substantially, because collisions can’t be prevented. Schedule-based transmission scheme works in a way that no collisions occur, for example, TDMA and FDMA. In this approach, resources such as time slots, frequencies, and spreading codes are explicitly allocated to network stations.

Normally, in schedule-based MAC, a centralized base station is required to make resource assignment for subscriber stations. However, over the past decade, a number of protocols have been proposed to provide distributed time-slot allocation. In REALM [2], once all the nodes in the two-hop neighborhood of a node have consistent information regarding their neighborhood, collisions are avoided because all the nodes use the same deterministic algorithm to select a winner for a given time slot.

A more efficient approach than REALM, is to let those nodes that were able to transmit during a prior frame notify their neighborhoods that they will not be competing for transmission slots for a period of time, which

effectively reduces contention among nodes and the delay experienced by a given node in accessing the channel [3]. For example, the Coordinated Distributed Scheduling (CDS) of IEEE 802.16 mesh mode introduces holdoff mechanism to reduce contention.

In this paper, we focus on CDS of 802.16 mesh mode for it is a comprehensive expression of up-to-date schedule-based WMN research.

Former researches on 802.16 CDS are mainly about adjusting holdoff time by considering node’s data queue length [4], number of neighbors[5], transmission state (sending, receiving, forwarding, or idle) [6], and whether to have three-way handshake information to send [7]. However, the SNR of different wireless channel is not taken into account, which will lead to inefficiency in time varying communication environment. Recently paper [8] begins to pay attention to that by adjusting modulation scheme to improve the band utilization for the mesh SSs with fine channel quality, but it doesn’t utilize dynamic holdoff time to control mesh contention.

Moreover, WMNs frequently act as the backhaul solution for a large number of access points serving a given coverage area. Therefore, the traffic loads among nodes of WMN is quite high. However, previous researches don’t pay much attention to this, and normally it is assumed that only partial links of WMNs are busy. For example, the simulation scenario is being devised that there are only 2 to 5 concurrent flows in the network, or the sending queue of a few busy nodes are not empty. When the backhaul WMN are fully occupied, the assumption of former research might be false. This results in failure of optimization.

To overcome the drawbacks of existing research mentioned above, in this paper, a distributed cross-layer adaptive scheduling algorithm (DCLASA) for WMN is proposed. The outline of this paper is as follows. Section II presents a survey of existing research on the coordinated distributed scheduling of WMN. Section III gives the system model and the algorithm description. Section IV shows the simulation results and a detailed analysis. Section V concludes the paper.

II. COORDINATED DISTRIBUTED SCHEDULING OF WMN

A. Mathimatical model of 802.16 mesh coordinated distributed scheduling In this section, we briefly review the 802.16 mesh CDS

from the three-way handshake perspective. The bandwidth

978-1-4577-0860-2/11/$26.00 ©2011 IEEE

distribution is done through the three-way handshake using request, grant and confirm information entity (IE). These IEs are carried within mesh distributed scheduling messages (MSH-DSCH).

In CDS, every node takes part in a collision-free contention, in which only a node within the two-hop neighborhood can win the next timeslot to transmit MSH-DSCH. And every node should inform its neighbors about the next MSH-DSCH transmission times of itself as well as its one-hop neighbors.

The exact time instant in which a node is able to transmit MSH-DSCH is called NextXmtTime. In order to save network resource and reduce the signaling overhead, mesh node do not broadcast NextXmtTime but XmtHoldoffExp and Mx, which defines the range of NextXmtTime.

2 < 2 ( + 1)XmtHoldoffExp XmtHoldoffExpMx NextXmtTime Mx⋅ ≤ ⋅ (1)

A node has to wait a time interval, which is called holdoff time (H), after its current MSH-DSCH transmission and before it starts to take part in the completion again, in order to share radio resource among mesh nodes fairly.

= 2XmtHoldoffExp baseH + (2)

In 802.16 standards, holdoff exponent is called XmtHoldoffExp, which is a constant ranging from 0 to 7, while base is fixed to 4. And Mx is calculated through formula (1).

M.Cao [9] studies the effect of changing holdoff exponent value and gives out the mathematical model and performance analysis of distributed scheduler in 802.16 mesh CDS. It brings forward that the MSH-DSCH transmission interval, which denotes the number of timeslots between successive MSH-DSCH, is a key factor that impact the efficiency of CDS.

Let τ denotes MSH-DSCH interval, then H Sτ = + (3)

While H denotes the holdoff time. S denotes the number of slots in which a node fails the contention before it wins, which is a random variable. The distribution of τ is sketched in Figure 1.

Figure 1. Interval τ between two successive transmissions [9]

MSH-DSCH interval is correlated with parameters such as holdoff exponent value, network topology, frame length, and the number of timeslots in a frame. Among all these parameters, the setting of holdoff exponent is the most flexible.

M.Cao suggests nodes with smaller holdoff exponents can have more chance to obtain data channel, thus network throughput can be improved.

However, too many nodes with small exponent value generate intensive competition that wastes system

resource. A good way to solve this is to adjust exponent values adaptively according to the competition node number variation to meet the connection QoS requirements.

B. Cross-layer design concept Cross-layer design refers to protocol design done by

actively exploiting the dependence between protocol layers [10]. Due to time varying characteristics of the wireless channel and the dynamic Quality of Service requirements, cross-layer design plays an important role in the next generation wireless systems.

In the research of 802.16 Mesh CDS, what is concern about cross-layer design is to consider QoS requirements of application layer, or the queue length of network layer, when designing scheduling algorithm in MAC layer. However, the difference in the qualities of the wireless channels is ignored, so the scheduling algorithm can’t adapt dynamically to the physical layer characteristics in order to improve the system performance.

III. SYSTEM MODEL & ALGORITHM DESCRIPTION

A. Background In this paper, we propose DCLASA, which is an

efficient, flexible and simple scheme for 802.16 mesh CDS. To improve throughput and reduce transmission delay, DCLASA adopts cross-layer design, which dynamically adjust the holdoff exponent in MAC scheduling module according to physical layer’s channel quality information in order to give users with good link SNR more opportunity to access time slots.

The theoretical foundations of DCLASA include adaptive modulation coding (AMC) and multi-user diversity. AMC is a link adaption technology in which the power of the transmitted signal is held constant over a frame interval and the modulation and coding format is changed to match the current received signal quality or channel conditions. Multi-user diversity (MUD) is obtained by opportunistic user scheduling, in which the transmitter selects the best user among candidate receivers according to the qualities of each channel between the transmitter and each receiver. Both of these technologies are used in wireless communication to improve system capacity, peak data rate, and coverage reliability.

B. Design Overview DCLASA is a mesh CDS algorithm, which take into

account wireless channel condition and other optimization factors. The main ideas of DCLASA are:

• Physical layer adopts AMC, which flexibly changes the modulation-coding scheme to the wireless channel condition, namely SNR.

• MAC layer adjusts scheduling algorithm according to the modulation-coding scheme decided by AMC. Because AMC is used, for a link, high data rates modulation scheme implies fine link quality. Therefore, according to MUD, MAC layer should adjust scheduling algorithm to increase the number of timeslots given to the node, in order to improve its bandwidth. On the contrary, node with low data rates modulation scheme implies low link quality.

Therefore, timeslots given to the node should be decreased to leave the chance to others.

• MAC layer readjust scheduling algorithm

according to other factors such as sending queue length.

C. System Framework Figure 2 shows the framework of DCLASA, which

consists of a mesh network adaptor including the physical and MAC layer. Compared with the common protocol stack, what is unique is the MAC scheduling module using DCLASA. It collect modulation scheme from AMC module and consider SNR is in accordance with the scheme. It might also collect other information such as sending queue length and user QoS requirements from other layers. After all valuable information is gathered, DCLASA use it to optimize timeslot assignment.

MAC layer

Physical player

AMC

Modulationscheme

SNR

Scheduling module using

DCLASA

Modulationschem

Figure 2. Framework of DCLASA

D. Approach Details Figure 3 is the flow chart illustrating sequential steps

for DCLASA. The terms used in the figure is defined as follows.

Figure 3. Flowchart of DCLASA

1) Active links

Active link refers to those links which have data to send or receive. In other words, these links have three-way handshake scheduling IEs to send, such as request IEs, grant IEs, and confirm IEs. Nodes which have active links should dynamically configure their holdoff exponents so as to give links with high data rates more timeslots.

2) Holdoff exponent mapping algorithm DCLASA dynamically sets holdoff exponent for each

node according to its modulation scheme. The holdoff exponent calculating formula is:

( ) ( ( -1- ) * ( -1) / ( -1) ) H x floor m x n m= (4) The formula’s input variable x stands for modulation

scheme of the link. Suppose there are m kinds of modulation scheme. [ ]0, m 1 and Nx x∈ − ∈ . In AMC, there are normally 7 to 9 modulation schemes, so [ ]7,9 m∈ . For example, AMC of 802.16d OFDM mode has 7 modulation schemes.

The output result H(x) stands for the value of holdoff exponent. Suppose n is the maximum value of holdoff exponent. ( ) [ ]0,n 1H x ∈ − . Considering the efficiency of transmission, normally n=5.

Floor(•) stands for the function of rounding up to the nearest integer.

The mapping of ( )x H x→ let m modulation indexes uniformly mapped to n holdoff exponent values, and make sure the modulation index denoting higher data rate is mapped to smaller holdoff exponent value.

The 5 values of holdoff exponent (n=5) stands for 5 QoS levels, which stand for different bandwidth requirements.

According to discussion above, we can infer that H(x) may map 2 different modulation indexes to the same holdoff exponent value, because [ ]0,m 1x∈ − and

( ) [ ]0,n 1H x ∈ − [ ] ( 7,9 and 5)m n∈ = . It is tolerable since 5 QoS levels can satisfy different service requirements, and no more precise level division is needed.

To simplify our simulation, let n=3 and m=4. Then the expression of H(x) is:

2 01 1

( )0 20 3

when xwhen x

H xwhen xwhen x

=⎧⎪ =⎪= ⎨ =⎪⎪ =⎩

(5)

As can be seen, H(x) maps 4 modulation indexes to 3 holdoff exponent values. 3) Compensation threshhold and counter

After node i derives its holdoff exponent value according to formula (4) or (5), the value is added to compensation counter c(i).

( ) ( )c i c i HoldoffExponent= + (6) The compensation threshold Cthresh is a fixed value.

When c(i)>=Cthresh , the holdoff exponent of node i needs to be set to zero so that it can take part in the timeslot contention immediately after current timeslot. Meanwhile, c(i) is reset to zero. In this way, the node with poor link quality can be periodically compensated to reduce its transmission interval so as to improve its throughput.

IV. PERFORMANCE EVALUATION In this section, we evaluate the performances of

DCLASA using NS2 simulation software. The 4*4 grid network topology, in which twenty-four traffic flows are generated from 16 nodes, is adopted for these performance studies. Considering the heavy traffic characteristics of WMN, we adopting the scenario that all links are busy sending and receiving, and the total offered loads of each traffic flows vary from 1Mbit/s to 24Mbit/s to investigate the performance variation from light load to heavy load. Each traffic source generates UDP data packets with the size of 1KB. The simulated time is 10 seconds. The radio propagation model is TwoRayGround. Table I shows the parameters used in our simulations.

TABLE I. SIMULATION PARAMETERS

Parameter Value

MSH-CTRL-LEN 8

MSH-DSCH-NUM 4

Modulation/Coding

scheme

QPSK 1/2, QPSK 3/4,

16-QAM 1/2, 16-QAM 3/4

Frame Duration 4ms

We compare the performance of our approaches (DCLASA-0 and 6) with the static approach in 802.16 mesh specification (EXP=0, 1, 2). DCLASA-6 denotes the compensation threshold is 6 and DCLASA-0 denotes never compensate. EXP=0, 1, 2 stands for the holdoff exponent is identical for each node and respectively equal to 0, 1, and 2. Figure 4 and 5 show the simulation results.

Figure 4. Network total throughput

Figure 5. End-to-end data packet delay

In figure 4 we can see that compared with static approaches, DCLASA has better network throughput. Particularly, DCLASA-6 has slightly improved the throughput of DCLASA-0, which means the compensation mechanism is in operation.

In figure 5 we can see that DCLASA also shows good performance in the simulation of end-to-end data packet delay. The packet delay of DCLASA (both 0 and 6) is lower than EXP=0 and 2. The reason why DCLASA has higher end-to-end delay than EXP=1 is that contention is more fierce in DCLASA than EXP=1. Therefore, a node might spend more time to win competition in DCLASA. But the throughput of EXP=1 is observably less than DCLASA. Considering all these factors, DCLASA outperforms the static approaches in 802.16.

V. CONCLUSION In this paper, we adopt cross-layer design to improve

system throughputs WMN by considering the channel quality differentiation in physical layer when designing the slot allocation algorithm in MAC layer, and adjusting holdoff exponent according to CQI. Based on this, a scheme named DCLASA is proposed for 802.16 WMN.

Former research on distributed scheduling algorithm doesn’t consider the wireless channel condition, so it can’t adapt to the wireless communication environment. However, DCLASA can solve this problem. It considers the heavy traffic characteristics of WMN by adopting the scenario that all links are busy sending and receiving with variable data rate. And considering cross-layer design between physical and MAC layer, it uses multi-user diversity and AMC technique to improve system efficiency.

Simulation by NS2 is done which proves that DCLASA can improve the network total throughput, and reduce link transmission delay. What is noteworthy is that DCLASA may also be applied to other wireless mesh networks after proper modification.

REFERENCES [1] Akyildiz I F, Wang Xudong, Wang Weilin. Wireless mesh

networks: a survey [J]. Computer Networks, Mar. 2005, Volume 47, Issue 4: 445-487

[2] D.Beyer, J.J. Garcia-Luna-Aceves, and C.Fullmer, Adaptive communication protocol for wireless networks, U.S.Patent Application, Feb.1999, Docket NO.003867.P001.

[3] BEYER D. System and method for collision-free transmission scheduling using neighborhood information and advertised transmission times, Patent WO/2003/019798, Jun. 2003.

[4] Loscri V. A queue Based dynamic approach for the Coordinated distributed scheduler of the IEEE 802.16 [C]. IEEE Symposium on Computers and Communications (ISCC), 2008: 423-428.

[5] Nico Bayer, Dmitry Sivchenko, Bangnan Xu, et.al. Transmission Timing of Signalling Messages in IEEE 802.16 based Mesh Networks [C], European Wireless 2006,Apr. 2006

[6] Nico Bayer, Bangnan Xu, Rakocevic V, et.al. Improving the Performance of the Distributed Scheduler in IEEE 802.16 Mesh Networks [C], IEEE 65th Vehicular Technology Conference, 2007.

[7] Liu Y, Feng S L, Ye W, et.al. A Dynamic Approach For Transmission Holdoff Time In IEEE 802.16 Mesh Networks [C]. 5th International Conference on Wireless Communications, Networking and Mobile Computing (WiCOM), 2009.

[8] Sun X.K. Differentiated service based cross-layer coordinated distributed scheduling algorithm for WMN [C]. International

Conference on Multimedia Information Networking and Security, 2009,vol. 2, pp.213-217

[9] Min Cao. Modelling and performance analysis of the distributed scheduler in IEEE 802.16 Mesh mode [C]. Proceedings of the 6th

ACM international symposium on Mobile ad hoc networking and computing, May 2005

[10] V. Srivastava, M. Motani, Cross-layer design: a survey and the road ahead [J], IEEE Communication Magazine, Volume 43, Issue 12, Dec. 2005 , pp.112 - 119