A Quota Transfer Protocol for Upstream Transmissions in Wireless Mesh Networks

Transcription

1 A Quota Transfer Protocol for Upstream Transmissions in Wireless Mesh Networks Yen-Bin Lee and Wen-Shyang Hwang Department of Electrical Engineering, National Kaohsiung University of Applied Sciences, No. 415, Chien Kung Road, Kaohsiung 807, Taiwan, R.O.C Abstract Wireless Mesh Network (WMN) makes people to pay much attention in recent years. In WMN, the data from subscriber can be relayed by multi-hop communication in the wireless mesh backhaul to the Mesh Portal Point (MPP) for accessing the wired Internet. This paper proposes a wireless scheduling protocol named Quota Transfer Protocol for data upstream transmission in wireless mesh backhaul. This protocol can guarantee the throughput from each mesh points to MPP in order to deal with the fairness problem and back-off inefficiencies caused by CSMA/CA mechanism. Keywords Medium access control, Wireless Mesh Network, quality-of-service(qos). I. Introduction By the development of wireless network recently, a commodity multihop Ad hoc Networks for Internet access technology which is named Mesh Networks make people to pay more attention on it [1]. As shown in Figure 1 WMN, the data from subscriber can pass by multi-hop communication through a row of immobile mesh points (MP) instituted backhaul to the Mesh Portal Point (MPP) for access the wired Internet, as shown in Figure 1. It benefits in extending the coverage areas, reduce cost and easy to deployment and maintenance. For the moment, several groups are working to define standards for wireless mesh networking techniques. The corresponding standards by range are IEEE a (WiMAX) for wireless metropolitan area networks (WMAN), IEEE s (Wi-Fi) for wireless local area networks (WLAN) and (ZigBee) for personal area networks (WPANS)[2]. Because WMN is a special case in mobile ad hoc networks (MANETs), there are many researches in routing and security in MANETs [3] which can be applied to WMN, we do not need to discuss these cases here. In this paper we only concerned about the bandwidth problem in the backhaul of WMN. Because of advancement of multimedia applications, the network Quality of Service (QoS) requirement which is included throughput, delay, and packet loss for multimedia transmission is more important. In WMN, take Wi-Fi for example, since all of the MP in backhaul use the same radio frequency, we must effectively arrange each MP transmitting data to avoid interference with other MP. If there is a mechanism for executing fairness for all MP sending data to the MPP, we can determine the nominal capacity of wireless mesh network [4]. A wireless token-passing protocol for multi-hop wireless mesh networks, named Ripple Protocol [5], proposed a downlink transmission mechanism for sending data from MPP to a leaf node. However, in uplink transmission, there will be a lot of MPs have data upstream to the MPP for accessing Internet. We must control each MP transmit data in order to deal with the fairness problem and avoid back-off inefficiencies. Figure 1 The network architecture of WMN We using ns-2 [6] to simulate the network architecture as shown in Figure 1 and explain the unfairness problem. We assume user 1 under MP N+4 generate UDP traffic (128kbps CBR flow 1) at 0.1 second of simulation time. At simulation time 3 seconds user 2 under MP N+3 generate a same UDP traffic (128kbps CBR flow 2). At 5 seconds user 3 under MP N+1 generate a higher UDP traffic rate at 512kbps (CBR flow 3). All the flows are toward to MPP. The simulation result as shown in Figure 2, flow 1 generate by user 1 with more hops, at first it can reach its throughput. But when flow 2 and flow 3 with fewer hops join and share the same backhaul, flow 1 gets dismal throughput and even starving. The unfairness is caused by 2 reasons, one is hidden and exposed terminal problems, and another is buffer management. We assume that there is an infinite amount of data to be sent from every node N+1 to N+4 in the network. In heavy loaded situation, when the data from node N+4 is sending to node N+3, it always dropped by node N+3 because the queue is full. Even we use different queues for the relayed and for the originating traffic and to serve them in a round-robin fashion [7]. The transmission probability of the data from node N+4 in node N+3 is 1/2. To analogize, the probability of data from node N+4 relayed by node N+2 is 1/4, relayed by node N+1 is only 1/8. However, the probability of data from node N+1 to MPP is 1/2, there still exists a serious fairness 338

2 problem. Besides, the mechanism under the contention base will cause the efficiency down. This paper proposes a mechanism to solve the problem on data upstream to MPP by a wireless scheduling protocol named Quota Transfer Protocol. This mechanism provides guarantee bandwidth approached to the nominal capacity of wireless mesh network for each MPs. Under this mechanism, if there are some MPs no data to be sent, the quota can send to other MPs as far as possible to reach highest utility rate. Figure 2 Simulation result - throughput of 3 flows II. Quota Transfer Protocol Overview By the routing algorithm such as AODV [3], each MP will record the routing table. This paper considers WMN architecture in a chain topology with a fixed routing path to MPP without discussing associated routing problems. As shown in Figure 3, the assumption of network architecture is similar to Ripple Protocol [5]. Each MP nodes are equally spaced and the radio of nodes does not interfere with other MPs except neighbors. In Figure 3, the node N is MPP, nodes from N+1 to N+4 are MPs. If each MPs always has data G to be sent; the collision domain corresponding to link node N+1 and node N+2 will be the max collision domain {N+3~N+2, N+2~N+1, N+1~N}. The max collision domain (MCD) = = 9. It means the nominal capacity G max <= Bandwidth / MCD = Bandwidth / 9. [4] The quota transfer protocol (QTP) proposed in this paper provides a guarantee bandwidth G QTP = Bandwidth / (MCD+1) to all MPs, which approaches to the nominal capacity G max under a heavy loaded situation. QTP scheduled MCD+1 timeslots to a Superframe (SF) as shown in figure 4. In a SF each MP node can send data to MPP. To achieve this objective, each MP must have two buffers and one [quota] counter. The first buffer is named [OB] (Output buffer) which is used to store the originating data from queue, and the other is named [RB] (Relay buffer) which is used to relay the data from downstream MPs. QTP adopts seven types frames: DATA, NULL, RTS, CTS, ACK defined in IEEE , Ready-to-Receive (RTR) frame defined in Ripple [5], and buffer Null and Ready-to-Receive (NRTR) frame defined in this paper. We use the same frame format defined in IEEE except the fixed duration DATA frame. The NRTR frame does not only include the same RTR function for DATA frame but also support the function to transfer quota. The QTP deals with RTS, RTR and NRTR as tokens for MP nodes to send data; MP node is only allowed to send its DATA frame when the quota is not empty and holds a RTS, RTR or NRTR. MP node can be operated in the one of the five following states: Initialization state: MPP send beacon and relayed to each MP. Every MP node gets quota value during this state. Transmit state (TX): After hold token, MP node is sending DATA or NULL frame to upstream MP node. Receive state (RX): MP node is receiving DATA or NULL frame from next MP node and store to [RB]. Listen state: Case MP node in Idle state overhear communication of other nodes will go to this state. MP node keeps silence to avoid the interference. Idle state: When errors occur or MP node expire its quota. We will use flowchart to describe of Quota transfer protocol shown in Figure 5. Figure 4 QTP Superframe Figure 3 The nominal capacity of a 4 nodes network in chain topology III. Illustrations of Quota Transfer Protocol This paper uses three scenarios to illustrate the quota transfer protocol. First scenario is in the heavy loaded situation; every MP node always has data to send. The second scenario is in the medium loaded situation; only part of MP nodes always has data to 339

3 send. The last scenario is a light loaded situation; there is only one MP node always has data to send. We assume the error rate during transmission is under control for example by using the FEC mechanism to correct errors. Scenarios 1: By the network architecture as shown in Figure 2, the wireless mesh backhaul has four mesh points from N+1 to N+4, and every MP node always has data to upstream send through MPP for accessing the wired Internet. Under this heavy loaded situation; the QTP shown as in Figure 6. In Figure 6, after beacon signal for synchronize all MP node, MP node N+1 and node N+4 first hold token and goes to TX state. MPP and MP N+3 received RTS goes to RX state. MP N+2 overhear RTS form MP N+1 and goes to listen state. In the first NAV time slot MPP received DATA from MP N+1. In the 4 th time slot MPP received DATA of MP N+4 relayed by backhaul. In the7 th time slot MPP received DATA of MP N+3. In the last time slot of SF, MPP received DATA of MP N+2. During this SF, each MP sent its [OB] data during one of NAV time slots, and all of the data relay hop by hop to the MPP. It shows QTP guarantees the throughput from each MP nodes to MPP. To complete a QTP process needs MCD+1 times NAV time slots, and the MCD = 9 from Figure 2; hence the upstream throughput G QTP can be got as follows: G QTP = Bandwidth/ (MCD+1) = Bandwidth/10 Figure 5 QTP flowchart Scenarios 2: Extending the network architecture of Figure 2 to six MP nodes from N+1 to N+6 and only four MP nodes {N+ 1, N+ 3, N+ 4, and N+ 6} always have data sent to MPP for accessing the wired Internet. Under the medium loaded situation, the QTP shown as in Figure 4. When the chain topology extends to six nodes, the collision domain still corresponding to link nodes N+1 and N+2 is the max collision domain {N+3~N+2, N+2~N+1, N+1~N}. The maximum of collision domain is: MCD = = 15. The SF after beacon is consisting of 16 time slots. It can get the upstream throughput G QTP as follows: G QTP = Bandwidth/ (MCD+1) = Bandwidth/16 In Figure 7, we verified a complete QTP process with sixteen NAV time slots. There is only MP nodes N+1, N+3, N+4 and N+6 have data to be sent upstream to the MPP. After a complete QTP process the MP nodes N+1 and N+4 sent one DATA frame and relay to MPP, that is, the MP nodes N+1 and N+4 have throughput G QTP. The MP nodes N+3 and N+6 sent two DATA frames and relay to the MPP; so the 340

4 MP nodes N+3 and N+6 get the throughput 2 G QTP. Every MP node with data to be sent will have throughput G QTP at least; that is to say, QTP guarantees the throughput G QTP for each MP node to MPP. In the protocol, MP node with no data also has to transfer its quota to others MPs for increasing the system efficiency and avoiding bandwidth waste. Figure 6 Timing diagram of Scenarios 1 under QTP mechanism Figure 7 Timing diagram of Scenarios 2 under QTP mechanism 341

5 Figure 8 Timing diagram of Scenarios 3 under QTP mechanism Scenarios 3: In the scenarios, the wireless mesh backhaul in Figure 2 has four mesh points from N+1 to N+4 but only MP node N+1 always has data sent to MPP for accessing the wired Internet. Under the light loaded situation; the QTP shown as in Figure 8. In Figure 8, MP nodes N+2 to N+4 have no data to send, but they are not in IDLE state. Under QTP these nodes still have to send NRTR and NULL for transferring its quota. As the result, the four quotas are all transferred to the MP node N+1, and the throughput of MP node N+1 is increased to 4 G QTP, and the system throughput is also grown. IV. Conclusions and Future Work This paper proposes a QTP under the chain topology of wireless mesh network architecture to provide a good schedule for MP nodes upstream transmitting to MPP. The protocol guarantees the throughput G QTP that approaches to the nominal capacity G max for each MP node sent to MPP in heavy loaded situation. It keeps the MP with more hops away from starving. When the load is not heavy, it transfers the unused quota to increase throughput as more as possible. That is, QTP provides the better QoS for the multimedia communication in throughput guarantee, delay, and packet loss. This paper is based on the architecture of wireless local area network (Wi-Fi); the extending this idea to the same architecture wireless mesh network in WMAN mesh (WiMAX) and PWAN mesh (ZigBee) will be studied in the future. REFERENCES [1] R. Bruno, M. Conti, and E. Gregori, Mesh networks: commodity multihop ad hoc networks, IEEE Commun. Mag., vol. 43, pp , Mar [2]Lee M. J.,Jianliang Zheng, Young-Bae Ko, Sherstha D. M., Emerging standards for wireless mesh Technology, IEEE Wireless Commun., vol. 13, pp56-63,april [3] Nikola Milanovic, Miroslaw Malek, Anthony Davidson, Veljko Milutinovic, Routing and Security in Mobile Ad Hoc Networks, IEEE Computer Society,pp.61-65, Feb [4]Jangeun Jun, Sichitiu M.L., The Nominal Capacity of Wireless Mesh Networks, IEEE Wireless Commun., vol 10,pp8-14, Oct [5]Ray-Guang Cheng, Cun-Yi Wang, Li-Hung Liao, Jen-Shun Yang, Ripple: A Wireless Token-Passing Protocol for Multi-Hop Wireless Mesh Networks, IEEE Communications Letters, vol 10, Feb [6] UCB/LBNL/VINT Network Simulator (NS-2), Available at [7]J.Jun and M. L. Sichitiu, Fairness and QoS in multihop wireless networks, in Proc. Of the IEEE Vehicular Technology Conference (VTC), October