An Efficient Backoff Algorithm for QoS Guaranteeing in Wireless Networks

Transcription

1 An Efficient Backoff Algorithm for QoS Guaranteeing in Wireless Networks Xinhua Liu, Guojun Ma, HaiLan Kuang, Fangmin Li School of Information Engineering, Wuhan University of Technology, Wuhan, {liuxinhua, maguojun, kuanghailan, Abstract: Wireless networks are required to guarantee the Quality of Service (QoS), including throughput, fairness, delay, etc. IEEE MAC protocol employs a typical backoff mechanism, Binary Exponential Backoff (BEB) algorithm, to avoid collisions when multiple users try to access the shared wireless channel. Many works on QoS guaranteeing point out that the BEB algorithm introduces the unfairness and serious delay problem into channel access. To solve the problem above, this paper proposes a novel backoff algorithm, which adopts a retransmission counter to measure the network congestion situation. The proposed algorithm also divides the contention window interval into small intervals, and in different intervals, it leverages different backoff strategies. Performance of this algorithm is compared with BEB and other improved backoff algorithms through simulation. The analysis and simulation results show that the proposed backoff algorithm can provide better QoS guaranteeing in terms of throughput, fairness, and delay. Key Words: Backoff algorithm, QoS, throughput, Fairness, Delay 1 INTRODUCTION Wireless networks are widely applied in industry control, transportation management, military reconnaissance, multi-media interaction, etc. In different application scenarios, different Quality of Service (QoS) requirements have to be achieved. In the military and emergency applications [1], messages need to be transmitted timely and accurately. Thus it requires that the network can guarantee short end-to-end delay and high transmission reliability. And in the multi-media interaction scenario, different users want to transmit enormous data at the same time. Thus it requires that the network can guarantee fairness and high throughput. Existing researches on QoS guaranteeing in wireless networks mainly focus on designing effective protocols in physical layer, data-link layer, and network layer [2-4]. Protocols in the MAC sublayer of data-link layer play an important role in controlling packet transmissions on the wireless channel. It is required to coordinate channel access with minimizing collisions. Hence whether the channel is allocated properly by the MAC layer influences the QoS guaranteeing in wireless networks. Existing MAC protocols are mostly based on IEEE Distributed Coordination Function (DCF) [5], which employs the Carrier Sense Multiple Access with Collision Avoidance (CSMA/CA) mechanism. If a node starts to transmit packets on the channel without noticing that another node happens to transmit packets on the channel at the same time, collision would occur, which leads to bad network situation. Using the CSMA/CA mechanism, a node has to listen to the wireless channel for a while to make sure whether the channel is already used by other nodes before transmitting This work was supported by the Natural Science Foundation of Hubei Province(No.2014CFB869) and the Fundamental Research Funds for the Central Universities (No.2014-IV-136 and 2013-IV-079). packets. If the channel is idle, the node is permitted to access the wireless channel and transmit packets. Otherwise, it has to defer a random time according to the Binary Exponential Backoff (BEB) algorithm. When a node unsuccessfully transmitted a packet, the waiting time for the node will get doubled in order to reduce the probability of collisions in next access turn. The BEB backoff algorithm is an effective strategy to reduce the probability of collisions. However, it introduces some problems [6], such as unfairness, delay, throughput, etc. What s more, it cannot properly adapt the contention window to different network congestion situations. Thus, in this paper, we propose a modified backoff algorithm to overcome the shortcomings of BEB algorithm. Network congestion situations have impacts on the probability of successful packet transmissions. Intense congestion situations might lead to frequently unsuccessful packet transmissions. In this case, it is hard for nodes to transmit packets successfully in one time, and nodes have to retransmit packets after unsuccessful packet transmissions. Hence the packet retransmission counter is adopted in the proposed backoff algorithm to measure the congestion situation in wireless networks. Backoff strategies change according to different congestion situations. Besides, the whole contention window interval is divided into small intervals, and different intervals adopt different contention window changing strategies. The proposed algorithm aims at guaranteeing QoS of throughput, fairness, and delay. Extensive simulations are carried out to evaluate its performance compared to BEB and other improved backoff algorithms. Simulation results show that our proposed algorithm can achieve better QoS guaranteeing in terms of throughput, fairness, and delay. The remainder of this paper is organized as follows. Section 2 studies existing researches on backoff algorithms, including BEB algorithm and some typical improved backoff algorithms. Considering the network congestion /16/$31.00 c 2016 IEEE 5353

2 situation, a new backoff algorithm is proposed in section 3, which employs the packet retransmission counter to measure the network congestion situation, and makes different backoff strategies in different situations. In Section 4 performance evaluations are conducted. And the proposed algorithm is compared with other exsiting backoff algorithms in terms of throughput, fairness, and delay. Section 5 concludes the paper. 2 RELATED WORK 2.1 IEEE DCF Backoff Mechanism: BEB Algorithm IEEE DCF is a fundamental technique of the MAC protocol based on WLAN standard, which employs the CSMA/CA mechanism with the BEB backoff algorithm [7]. When a node wants to transmit packets, it defers for a random time according to the BEB algorithm, which forces the node to access the wireless channel after a waiting time. Multiple users choose the backoff time independently. More than one node may choose the same backoff time. Thus it decreases the probability of collisions. The length of the backoff time is shown as the following equation: Tbackoff = N Tslot (1) where T backoff is the actual backoff time, and N is a random integer, uniformly distributed within the interval (0, CW); CW denotes the contention window, which is between the minimum contention window CW min and the maximum contention window CW max; and T slot is the slot time de ned as 20 s by the physical layer. The BEB algorithm is the key point of the DCF mechanism, which can be briefly expressed as follows: when a node successfully transmits packets, the CW of the node will reset as the minimum contention window CW min; and if a collision happens or a node unsuccessfully transmits packets, the CW will get doubled. The value of CW can be calculated as follows: CWfailure = min( CW 2, CWmax) (2) CWsuccess = CWmin The value of CW, which varies in a certain range, can measure the ability of a node accessing the channel. The smaller the CW is, the shorter the backoff time is, and the higher the probability of a node successfully accessing the channel is. Since different nodes have different random backoff time, they can get into the channel with low probability of collisions. 2.2 BEB Shortcomings and Improved Algorithms The BEB algorithm has two shortcomings. First, it introduces unfairness problem [8]. As described in the last subsection, the node, which transmits packets successfully, resets its CW as the minimum contention window CW min, and other nodes, which failed to transmit packets or unsuccessfully access the channel, have to double their CW. In the next turn of contention, the node with smallest CW has the highest probability in accessing the channel, leading to a fairness problem. The fairness problem is caused by the drastic change of the CW, thus an alleviating method is to keep the CW change smoothly. What s more, when the packet is successfully transmitted, the CW will decrease to CW min, and the congestion situation will not be reflected properly. When the wireless channel is busy and collisions occur frequently, the CW of each failed node increases quickly. In this case, if a node transmits a packet successfully, its CW will reset to the CW min immediately, which brings the node an illusion that the channel is not busy now. In short, the BEB algorithm cannot reflect the channel situation. To overcome the shortcomings of the BEB algorithm, some improved algorithms are studied. The Slow Contention Window decrease (SD) algorithm [9] adopts the multiplicative decrease mechanism. When a node transmits a packet successfully, the CW decreases as half as before. SD algorithm will enhance the abilities of guaranteeing throughput and fairness, and it may introduce intolerant delay problem occasionally. The Multiplicative Increase and Linear Decrease (MILD) [10] algorithm employs gradual CW changing strategy against the drastic contention window changing strategy in BEB algorithm. When the number of active nodes in the network is large, the performance of MILD is much better than BEB. When the number of the nodes is small, the CW decreases slowly due to the linear decreasing, thus the extra delay problem is introduced. The Multiplicative Increase and Multiplicative/Linear Decrease (MIMLD) [11] algorithm adopts a multiplicative increase and multiplicative and linear decrease strategy in order to be adaptive to current network conditions. MIMLD algorithm adopts a new mechanism for the initial contention window choice. It has two initial contention windows named CW min and CW basic, and it chooses the initial contention window adaptively between CW min and CW basic. It can improve the performance of fairness and end-to-end delay. However, it cannot adapt itself to the network situation variations in time. Considering the changes of the network situation, a new adaptive backoff algorithm is proposed in this paper. 3 AN EFFICIENT BACKOFF ALGORITHM Our proposed algorithm adopts a boundary control parameter CW mid, dividing the whole contention interval into two small intervals (CW min, CW mid) and (CW mid, CW max), where CW min, CW mid and CW max are all constant integers as listed in Table I. (CW min, CW mid) is referred to as the light traffic load interval, and (CW mid, CW max) is the heavy traffic load interval. Network congestion situations have impacts on the probability of successful packet transmissions. Hence the packet retransmission counter is introduced in the proposed backoff algorithm to measure the congestion situation. The modified algorithm employs different CW changing strategies according to different traffic load intervals and retransmission counts. The algorithm is called as Retransmission Count Adaptive Backoff (RCAB) algorithm. The flow of RCAB algorithm is described as shown in Figure 1. In the RCAB algorithm, if the CW is located in the interval (CW min, CW mid), it supposes that the network contention is not intense. Hence, if a node failed to transmit a packet, the th Chinese Control and Decision Conference (CCDC)

3 CW will increase by a times. The value of a is chose in the range of from 1 to 2, and the best value of a is 1.8 through simulations. Thus the CW does not increase too fast when the traffic load is light. If a packet is transmitted successfully, it will subtract b from the current CW in order to keep the CW changing smoothly. And b is defined as the integer number 1 here. Fig. 1. The flow of RCAB Algorithm On the other hand, if the CW is located in (CW mid, CW max), the network contention now is intense. And we employ the packet retransmission counter to measure how intense the network congestion is. If the packet retransmission happens and retransmission times are fewer than Cr (Cr is the boundary times defined as the integer 4), the CW will increase by c times (c is defined as 1.5), where c is fewer than a, since the CW in the case of intense network contention is not expected to increase too fast. What s more, if the retransmission time is more than Cr, it suggests that the network contention is intense. The CW has to change more smoothly. And the CW adopts the linear increasing strategy instead of multiplicative increase strategy. It will add d (d is defined as the integer 2) to the current CW in order to make the CW increase smoothly. If transmission succeeds, the CW will be divided by e times (e is defined as 1.4), and e is a little smaller than c. If the packet retransmission count is more than the specified maximal retransmission time defined in physical layer, the CW will be reset as the initial value CW min. 4 PERFORMANCE EVALUATION In this section, we carry out simulations on NS-2.34[12], a discrete event simulator targeting at network research supported by UC-Berkeley University. The simulation scenario is built as follows to analyze and compare the RCAB algorithm with BEB and some other typical improved backoff algorithms mentioned before. Fig. 2. Simulation topology As shown in Figure 2, there are 4 nodes in a line, which are labeled Node 0, Node 1, Node 2 and Node 3. Every two neighbor nodes are separated 150 m. The transmission distance is 250 m. The solid lines stand for the traffic links. We put two traffic flows on the two solid lines at the same time. Flow 1 is put on the link from Node 0 to Node 1, and Flow 2 is put on the link from Node 2 to Node 3. The two senders (Node 0 and Node 2) are out of communication range of each other, but the receiver Node 1 of Flow 1 and sender Node 2 of Flow 2 are in the communication range of each other. If Node 0 is sending packets to Node 1 and Node 2 has some data to transmit to Node 3, Node 2 has to carry out the carrier sense first. As the Node 2 is in the communication range of Node 1, Node 2 will hear the transmission from Node 0 to Node 1, and learn that the channel is not available. Node 2 is prevented from transmitting messages to Node 3. Hence the hidden node problem occurs. As shown in Figure 2, Node 0 and Node 2 are hidden nodes to each other. The problem of hidden nodes can reduce the capacity of network because of the waste of the wireless channel. We evaluate the performance of the proposed algorithm in terms of throughput, fairness and delay in following subsections. And corresponding simulation parameters are listed in Table 1. Table 1. Simulation Parameters Parameter Value Parameter Value Slot Time 20 s CTS 304 bit SIFS Time 10 s ACK 304 bit DIFS Time 50 s CW min CW max RTS Threshold 3 kbit BEB PHY Header 192 bit SD MAC Header 224 bit MILD CW Duration 300 s CW min/ CW basic MIMLD RTS 352 bit 2/ CW max Packet Size 1 kbit CW min CW mid CW max RCAB Duration 100 s Throughput Comparison We put on two traffic flows at the same time in the network, and conduct simulations in NS2 to see throughput performances of two flows using the RCAB algorithm, the BEB algorithm and some other typical improved backoff algorithms. Figure 3 presents the throughput of the two different flows using BEB algorithm. We can see that the throughput of flow 2 grows with the increasing of the traffic load until the maximal throughput 710 kbps while the throughput of Flow 1 is restrained. It decreases with the increasing of the traffic th Chinese Control and Decision Conference (CCDC) 5355

4 load. The simulation results show that different traffic flows cannot get nearly equal throughput due to BEB algorithm. Figure 4 shows that the throughput of RCAB algorithm increases with the increasing of the traffic load until the max throughput about 800 kbps. From Figure 3 and Figure 4, we can see that when the traffic load is heavy (rate > 400kbps), the two different flows of RCAB algorithm get nearly equal throughput while the throughput of different flows of BEB algorithm is far away from each other. with the increasing of the traffic load until the maximal throughput. Fig. 5. Throughput of the total flows with BEB and RCAB algorithms Fig. 3. Throughput of two traffic flows of BEB algorithm Fig. 6. Throughput of the two traffic flows with different backoff algorithms Fig.4. Throughput of two traffic flows with RCAB We compare the total throughput between BEB and RCAB algorithm. From Figure 5, we can see that when the traffic load is light, the two different algorithms get nearly equal performance in terms of throughput. And when the network is in the state of heavy traffic load (for example, rate = 900 kbps), the RCAB algorithm reaches twice of the total throughput of BEB algorithm. So, the RCAB algorithm has better capacity of guaranteeing throughput than that of BEB algorithm. Figure 6 presents the throughput of two flows with different backoff algorithms, including SD algorithm, MILD algorithm, and MIMLD algorithm, which are mentioned in section II. This will be used to compare the fairness of the RCAB algorithm with existing typical improved backoff algorithms in the next subsection. We also find that the throughput of those improved backoff algorithm increases 4.2 Fairness Comparison In this subsection, we compare the fairness among different backoff algorithms. And fairness is the most important parameter in our algorithm. We employ the Fairness Index (FI), referred to reference [13], to measure the fairness performance. We denote { TH i i = 1, 2,, N } as the throughput of the ith link in the network, and N is the total number of links. The formula FI is defined as follows: TH max FI = (3) TH min Where TH max = max( TH i ), TH min = min( TH i ). i i The more the FI is, the more unfair for nodes to access the channel. When the value of FI is 1 or close to 1, the network achieves good performance in fairness. Figure 7 shows the fairness comparison between BEB and RCAB algorithm. When the traffic load of network is light (rate < 400bps), the two algorithms have almost the same th Chinese Control and Decision Conference (CCDC)

5 performance on fairness, but when the load is heavy (rate > 400 kbps), the RCAB algorithm gets better fairness performance than the BEB algorithm. The fairness comparison are carried out between RCAB algorithm and some typical improved backoff algorithms, using data as shown in Figure 6. And the result as shown in Figure 8 shows that RCAB algorithm has better capability of fairness guaranteeing than other backoff algorithms especially when the traffic load is heavy. We can also find that MILD algorithm achieves as good fairness performance as RCAB algorithm gets. So we compare the fairness between RCAB and MILD in different scenarios. algorithm. It can be concluded that the RCAB algorithm has better capability of fairness than other algorithms mentioned above. Fig. 9. Simulation topology for the FI comparison of MILD and RCAB Fig.7. Fairness performance with BEB and RCAB Fig. 8. Fairness performance with RCAB and other improved algorithms Simulation experience on different topologies is carried out to verify the validity of RCAB. There are two cells in Figure 9. Each cell contains a traffic flow. In cell C1, a traffic flow is built from Node 0 to Node 2. And in cell C2, a traffic flow is built from Node 5 to Node 7. The distance between two neighboring nodes is 50 m. As the transmission range is 250 m, all the nodes in cells C1 and C2 are in communication range of each other. We compare the fairness performance between MILD and RCAB. The result as shown in Figure 10 indicates that the RCAB algorithm achieves better fairness capability than the MILD Fig. 10. Fairness performance between MILD and RCAB 4.3 End-to-end Delay Comparison In this subsection, we carry out the end-to-end delay comparison among different backoff algorithms. The end-to-end delay of two thousand packets are recorded. Figure 11 shows that the end-to-end delay of the new RCAB algorithm is about 1.0 s while that of the BEB algorithm is about 2.5 s on average, so the performance of delay of RCAB is improved by 60% than that of BEB. What s more, the delay jitter of RCAB is much smaller that of BEB. We can also find that the end-to-end delay of the SD algorithm is worse than that of the BEB algorithm, since the SD algorithm adopts the fast increasing strategy when the packet transmission is failed and low decreasing mechanism when packet transmission is successful in order to get better fairness performance, ignoring the delay guaranteeing problem. The MILD algorithm gets similar results with the SD algorithm, and the jitter of the end-to-end delay changes more smoothly as the simulation goes on. The SD and MILD algorithm get better fairness guaranteeing at the cost of end-to-end delay guaranteeing. As shown in Figure 12, the capacity of delay guaranteeing of RCAB algorithm is improved by about 10% compared with MIMLD algorithm. The distribution of the packet of RCAB algorithm is more concentrative than that of MIMLD algorithm. Finally, we make a summary on the performance comparison with different backoff algorithms according to these simulation results above. Table shows us that our RCAB alogrithm, compared with BEB algorithm and other three types of typically backoff th Chinese Control and Decision Conference (CCDC) 5357

6 algorithms, can achieve better performance in terms of throughput, fairness, and dely. backoff algorithm. From our simulation results, it is observed that the proposed algorithm achieves better throughput, fairness and delay performance than that of BEB algorithm and other three types of typical backoff algorithms. REFERENCES Fig.11. End-to-end delay with different algorithms Fig.12. End-to-end delay with RCAB and MIMLD Backoff Algorithm TABLE.PERFORMANCE COMPARISON Throughput Fairness Delay BEB Bad Worst Bad SD Good Bad Bad MILD Good Good Bad MIMLD Good Bad Good RCAB Good Best Best 5 CONCLUSION In this paper, QoS guaranteeing of fairness, throughput and delay is achieved by proposing an efficient adaptive backoff algorithm based on IEEE MAC protocol. The proposed backoff algorithm employs the retransmission counter to measure the network status and divides the whole contention window interval into two small intervals. And in different small contention window interval, different backoff strategies are applied. We carry out related analysis and simulations to evaluate the performance of the proposed [1] Y. Gadallah and M. Serhani. A WSN-driven service discovery technique for disaster recovery using mobile ad hoc networks. Wireless Days (WD), 2011, pp. 1-5 [2] T. Duong, H. Zepernick, and M. Fiedler, Cross-Layer Design for Integrated Mobile Multimedia Networks with Strict Priority Traffic. IEEE Wireless Communications and Networking Conference (WCNC), 2010, pp. 1-6 [3] Young-Ae Jeon, Sang-Sung Choi, Sung-Woo Park and Dae-Young Kim. QoS routing for distributed MAC-based high rate wireless personal area networks. 5th International Conference on Computer Sciences and Convergence Information Technology (ICCIT), 2010, pp [4] J. Ben-Othman, L. Mokdad, and B.Yahya. An Energy Efficient Priority-Based QoS MAC Protocol for Wireless Sensor Networks. IEEE International Conference on Communications (ICC), 2011, pp. 1-6 [5] Editors of IEEE Wireless LAN Medium Access Control (MAC) and Physical Layer (PHY) speci cations, Draft Standard. IEEE , 1997 [6] S.Xu and T. Saadawi. Does the IEEE MAC protocol work well in multihop wireless ad hoc networks? IEEE Communicaitons Magazine, 2011, pp [7] Byung-Jae Kwak, Nah-Oak Song and L. Miller. Performance analysis of exponential backoff. IEEE/ACM Transactions on Networking, 2005, pp [8] K. Medepalli and F.A. Tobagi. On Optimization of CSMA/CA based Wireless LANs: Part I-Impact of Exponential Backoff. IEEE Internaltional Conference on Communications (ICC), 2006, pp [9] A. Ksentini, A. Nafaa, A. Gueroui and M. Naimi. Determinist contention window algorithm for IEEE IEEE 16th International Symposium on Personal, Indoor and Mobile Radio Communications, 2005, pp [10] Nah-Oak Song, Byung-Jae Kwak, J. Song and M. Miller. Enhancement of IEEE distributed coordination function with exponential increase exponential decrease backoff algorithm. IEEE 57th Semiannual Vehicular Technology Conference, 2003, pp [11] Vaduvur Bharghavan, Alan Demers, Scott Shenker. MACAW: a media access protocol for wireless LANs. Proceedings of the ACM SIGCOMM Conference on Communications Archiotctures, Protocols and Applications,1994, pp [12] [13] T.Ozugur, M.Naghshineh, P.Kermani, and J. Copeland. Fair media access for wireless LANs. IEEE Global Telecommunications Conference (GLOBECOM), 1999, pp th Chinese Control and Decision Conference (CCDC)