A Generic Mobile Node Architecture for Multi-Interface Heterogenous Wireless Link Layer

Transcription

1 A Generic Mobile Node Architecture for Multi-Interface Heterogenous Wireless Link Layer Rafaa TAHAR HANA Research Group University of Manouba, Tunisia Abdelfettah BELGHITH HANA Research Group University of Manouba, Tunisia Rafik BRAHAM Prince Research Unit University of Sousse, Tunisia Abstract In this paper, we first propose a new and open architecture for mobile wireless nodes using multiple interfaces. This modular and flexible architecture aims to offer to industrial and research communities several possibilities to investigate and simulate multiple wireless protocol aspects, algorithms and issues. To validate our architecture and to show its advantages, we choose to implement a multiple interface WLAN protocol in the widely used OMNeT++ Discrete Event Simulator [1] based on the MMAC Protocol [2] and its derivatives [3]. While using multiple interfaces leads certainly to an enhancement of the entire network performance, the question remains to specify algorithms and mechanisms that realize a compromise between power consumption and performance efficiency (Throughput, End-to-End Delay). Secondly, we present a modified version of the PSM-MMAC Protocol [3] that we called the Adaptive ATIM Multi-Interface MAC Protocol (AA-MIMP). Using our proposed open architecture, we show that AA-MIMP realizes a substantial enhancement of QoS parameters and performances, yet demonstrating a substantial improvement of power consumption. I. INTRODUCTION /12/$31.00 c 2012 IEEE In the last few years the deployment of wireless technologies, especially WLANs based on the IEEE standard [4], [5], has grown exponentially. However, this technology still suffers from certain inefficiencies such as low throughput, inefficient power conservation, lack of real time consideration for peer-to-peer and interactive applications including VoIP, VoD, Streaming, etc. Several recent research works tried to propose enhancements and thrived to offer some solutions. Of particular interest to our present work are those based on the efficient use of the available frequency spectrum instead of just being limited to a unique radio channel as currently deployed. The major contribution of this work is to propose a new architecture of a mobile node that permits the design and implementation of algorithms and protocols allowing the enhancement of wireless communication. We also demonstrate how the new architecture could implement a multi-interface protocol for IEEE WLANs. This paper is organized as follow. In the first section we present details and key features of the new architecture. In the second section we show how to implement a modified version of the PSM-MMAC Protocol [3] that we call Adaptive ATIM Multi-Interface MAC Protocol (AA-MIMP). In the last section we will conduct a performance evaluation to show the enhancement of QoS parameters (throughput and end-to-end data delay) and a substantial improvement of power consumption. II. MOTIVATIONS AND RELATED WORK Many works dealt with the multi-channel / multi-interface wireless communications [6], [7], [8]. Some of them proposed to set up virtualization techniques at the Link layer (MAC and LLC) [9]. These techniques aim to maintain specific data structures in order to realize suitable mapping between the logical information of the Network Layer and the physical information of the Link and Physical layers. Some works deal with multi-channel / multi-interface heterogeneous wireless communications issues, but to the best of our knowledge, only the Hyacinth [10] architecture tried to propose a multi channel hybrid technology IEEE /IEEE [11] integration. The Hyacinth architecture proposes to use multiple IEEE channels to define aggregate coverage areas for mobile nodes and uses IEEE or IEEE technologies to ensure the connectivity of these areas to wired network. But this architecture did not deal with nodes that combine multiple technologies which is the case nowadays with mobile nodes having multiple built-in wireless technologies: IEEE , IEEE p [12] and IEEE 1609.x (VANET), IEEE [13],... In this paper, we will focus solely on the multi-interface approach for several reasons. The NIC (Network Interface Card) has a low cost and still decreasing. This approach offers more flexibility in implementation and a relative relaxation from the constrained aspects of synchronization between clocks. Research and implementations already conducted at HANA Research Group showed the advantages of the Multi-interface approach [14], [15], [16]. III. PROPOSED MULTI INTERFACE NODE ARCHITECTURE FOR WIRELESS NETWORKS In order to take advantage of the available frequency spectrum within a wireless network, and regarding the low cost of wireless NIC, a simple idea consists of equipping each mobile node with multiple NICs. But this NIC multiplicity will induce many complex issues. The most important of them are : NIC management (especially for heterogeneous NICs), use of multiple wireless technologies, overhead optimization, power conservation, clocks drift, interface selection and traffic

2 distribution. To solve these problems, we propose a new mobile node architecture that must also guarantee that protocol modifications should be minimal and flexible to authorize future works for both industrial and research communities. The key features of our model could be resumed in these points: The implementation of a Layer-2 virtualization technique called VML (Virtual MAC Layer) by derivation/overloading of the LLC sub-layer. Indeed this VML will ensure an abstraction of the NIC multiplicity regarding upper layers (IP and upper layers); An interface is represented by a PM-Bloc composed of a Physical and MAC layer which makes our architecture physically aware (no modification of the Physical Layer); We implement necessary but not exhaustive data structures and algorithms for interface selection policies, NIC status identification (sender, receiver, inactive); Our implementation is integrated and tested within the widely used OMNeT++ Discrete Event Simulator v4.1. A. Multi Interface Mobile Node Architecture In mostly used network discrete event simulators (NS-2, J-SIM, OMNeT++,...), a classic mobile node is composed of : A Physical layer, A MAC sub-layer that implements the protocol state machine, A Management sub-layer (LLC) that implements management procedures, complementary modules that implement wireless communications specific aspects such as : mobility, propagation models, cross-layering features,... and upper layers modules (IP, Transport, Applications). When we deal with efficient use of different orthogonal wireless channels there are mainly two modes : multi-channel and multi-interface. The difference between these two modes could be resumed as follows: Multi-interface mode allows the simultaneous use of the total frequency spectrum Multi-channel mode reduces channel noise and interference Since every interface is tuned to a given channel, multiinterface MAC layer management is more flexible and easy because it does not need channel switch, scan,... Multi-interface LLC layer management is more complex since other procedures are necessary such as : multiple MAC address management, power conservation,... Since multi-interface mode is more efficient, we conducted a complete study and we propose a new mobile node architecture, as shown in Figure 1. It is composed of : PM-BLOCs that associate a MAC sub-layer state machine and a physical module A VML (Virtual MAC Layer) that substitute the LLC sub-layer. Notice that our architecture could also be generalized to cover multi-channel issues by introducing switching time between channels and activating only one PM-Bloc at a time. Fig. 1: New Mobile Node Architecture 1) Physical and MAC Sub-Layer BLOC (PM-Bloc): The hard coupling between the physical layer and the MAC sublayer is due to two objective constraints. They are : MAC sub-layer is a logical representation of the physical underlying layer by means of MAC protocol state machine; There is no more distinction between the modules since in the multi-interface mode each NIC is tuned to a specific channel. Recall that since we choose to implement the PSM-MMAC protocol, some modifications are introduced to the control PM- Bloc. Other blocs are left as-is. 2) VML : Virtual MAC Layer: To handle the multiplicity of NICs (PM-Blocs), it is necessary to derive/overload the LLC sub-layer in order to deal with: The abstraction (virtualization) of multiple NICs regarding upper layers; Multiple NICs management (Management Frames like ATIM, channels identification, PSM,...); Traffic distribution. Our new architecture which combines VML and PM-Blocs to describe a multi-interface mobile node could be used to study multiple scenarios such as Wireless Distribution Systems (WDS) [15], [16], multi-interface Ad-hoc WLANs and VANET. As an application of this architecture, we choose to implement and enhance the PSM-MMAC protocol [3] based on both the MMAC [2] and the DCA protocol [17]. Since this protocol family is based on the idea that all nodes must at least have 2 NICs (one dedicated to control and the others are for data), we introduced solely some modifications to the control PM-Bloc. Data PM-blocs are not modified. IV. ADAPTIVE ATIM MULTI-INTERFACE MAC PROTOCOL (AA-MIMP) We now propose an improved version of the PSM-MMAC in the quest to enhance power saving and improve overall performances. We implement the AA-MIMP using the proposed new architecture. The AA-MIMP enhances the PSM-MMAC protocol [3] and is based on both the MMAC [2] and the DCA protocol [17] in order to enhance QoS parameters and

3 to reduce power consumption. The AA-MIMP is dedicated to WLAN in which nodes are in Line of Sight (LoS) and could be heterogeneous (number of NICs varies from a node to another). Unlike the PSM-MMAC Protocol that uses a complex probabilistic model to estimate the size of the ATIM Window, our protocol is based on the standard DCF function and uses control channel sensing to decide or not if the ATIM / ATIM-ACK Frames exchange is terminated or not. Data structures maintained and information exchanged between the control PM-Bloc and VML allow us to determine the exact state of all the network during the current Beacon Interval in a fully distributed manner. Like the DCA Protocol, our protocol is receiver side channel selection oriented but does not add the RES frame to indicate to neighbors which channel is selected for the transmission. Instead, it piggybacks this decision in the ATIM-ACK frame. Our Protocol is based on the standard timing structure proposed in the PSM and tries to solve the following problems : Dynamic ATIM Window Size estimation regarding the submitted traffic; Interface selection regarding a traffic distribution specific policy (random, load balancing, threshold,...); QoS parameters enhancement; Power conservation. Unlike the complex probabilistic model for ATIM Window Size Estimation proposed in the PSM-MMAC Protocol based on Active Links (AL) and which suppose that traffic loads submitted to the network are constant, our estimation algorithm is most adaptive and is based on the observation of Packet s Queue Size at each start of a Beacon Interval and we did not make any assumption on traffic loads which is the case of WLANs, WDS,... So our algorithm is based on the following key features : At the start of every Beacon Interval, all control PM- Blocs are awake and still awake for at least the rest of the ATIM Window; Only nodes that have traffic will send an ATIM Frame using the DCF function. So in the ATIM Frame body a mobile node will send the following information: Packet Number (actual Queue Size), Mean Packet Size and available Data Channel List; Receiver node selects the data channel regarding a predefined selection policy (common for all nodes during the current Beacon Interval, in our case a load balancing policy) and transmits the selected channel number in the ATIM-ACK Frame from the common subset channels with the sender. The choice of this receiver oriented selection policy is due to the fact that sender and receiver mobile nodes may have heterogeneous capabilities; When the sender receives the ATIM-ACK Frame it will immediately send its traffic on the dictated data channel. If the dictated channel is the control PM-Bloc, the sender must delay transmission until the end of the ATIM Window Estimation Algorithm; Estimated ATIM Window Size is decided when an amount of slot time equal to the upper bound of the size of the current contention interval, that we have called backof f F actor, without sensing any activity on the control channel or when the bound maximal value of the ATIM Interval, as mentioned in the PSM standard equal to 0.02 sec, is reached; At the end of the ATIM Estimated period, all nodes know exactly the status of every channel and which PM-Blocs are active and those which must go into doze mode. V. PERFORMANCE EVALUATION A. Definitions and notations We now discuss the AA-MIMP QoS parameters, namely the throughput and the end-to-end delay. We also discuss Duty Cycle and Per bit Power Consumption formally defined as follows : Throughput : The quantity of information per time unit delivered successfully to all destination stations. That is : nbp ckt pcktsize T hroughput = (1) simduration where nbpckt denotes the number of packets received successfully by all destination stations. pcktsize denotes the size of a packet and simduration denotes the simulation duration in seconds. End-to-end delay : The average end-to-end delay of all packets delivered successfully to all destination stations. That is : Delay = DestArriveT ime SrcGenerationT ime (2) where DestArriveTime denotes the time at which the packet arrives at the destination station and SrcGenerationTime denotes the generation time of this packet at the source station. Duty Cycle : We define duty cycle as the average amount of unused transmission time periods of active channels per beacon interval. That is: dutycycle = BeaconT ime() LastP ackett ime() (3) where BeaconTime() denotes the current beacon time and LastPacketTime() denotes the time at which the last DATA or ACK packet is sent or received. Per bit Power Consumption : We define it as the average quantity of energy consumed by all node channels divided by the network throughput. We assume that a node has four energy states: Idle (1.0 W), transmission (1.8 W), receive (1.3 W) and doze (0.05 W). B. Simulation scenario Table I presents general parameters for our simulation.

4 TABLE I: General Parametrs Parameter PYH and MAC bitrate Data channel number Beacon Interval ATIM Max Value Flow Number Packet Size Packet tic Simulation Time Transmission range Value 2 Mbps 1, 2 and 4 channels 0.1 sec 0.02 sec 1, 5, 10, 15 and 20 Flows 512 Bytes 0.01 sec 200 sec (2000 Beacon Periods) 300 meters Fig. 4: Throughput vs. Flow Number phenomenon in the PSM-MMAC protocol but for a number of flows equal or higher than 4. Fig. 2: Estimated ATIM Window Size vs. Flow Number E. ATIM Window Size Estimation vs. Throughput To validate our ATIM Window Size estimation regarding the throughput parameter we measure the Throughput function of ATIM Window Size. So we will consider 10 flows with 1 data channel (10 packets per beacon interval) and we will vary the ATIM Window Size from sec to sec. As we can C. ATIM Window Size Estimation Figure 2 illustrates the Estimated ATIM Window size vs. Number of flows and shows that our estimation is independent of the number of data channels. The figure Fig 3 shows the number of ATIM / ATIM-ACK frame exchange succeeded during the ATIM Interval. This number decreases while number of flows increases which is due to high control channel contention. Fig. 5: Throughput vs. ATIM Window Size see in the Figure 5, the maximal throughput is obtained for an ATIM Window Size value equal to sec, which is the value obtained by our estimation as shown by Figure 2. F. End-to-End Delay parameter Figure 6 shows the End-To-End DATA Delay vs. Number of flows. We notice that with 4 interfaces, End-To-End DATA Fig. 3: Correctly received ATIM D. Throughput Parameter Figure 4 shows Throughput vs. Number of flows. By comparison with the PSM-MMAC protocol we notice that our protocol gives better performance with 1 and 2 data channels for 10 flows and 10 packets/beacon interval and same performance with 4 data channels. We notice that when the number of flows is equal or higher than 15, a degradation is noticed. This could be explained by both the increasing number of unsuccessful ATIM / ATIM-ACK frame exchange, as shown by Figure 3, and DATA Collision number that increases due to large channel contention.we notice the same Fig. 6: End-to-End Delay vs. Flow Number Delay is less than 100 ms for 10 flows (exactly ms). This means that multimedia communications are possible. Figure 7 is a zoom for 1 and 5 flows for more clarity.

5 2 data channels. For a high network load (number of flows greater than or equal to 15), energy cost per bit is minimal for 4 data channels. This means that our intelligent power consumption approach realizes a good compromise between QoS parameters and power consumption and outperforms the PSM-MMAC Protocol. Fig. 7: End-to-End Delay vs. Flow Number (Zoom) G. Duty Cycle Parameter Figure 8 shows the duty cycle vs. the flow number. We VI. CONCLUSION Adaptive ATIM Multi-Interface MAC Protocol (AA-MIMP) shows that our new architecture of mobile node is valid, flexible and open to other additions and improvements. Our Protocol realizes a good compromise between QoS parameters (Network Throughput and End-to-End DATA Delay) and power consumption by intelligent turning on used and off unused interfaces. This result is confirmed by extensive simulations which show that the AA-MIMP outperforms the PSM-MMAC protocol in terms of QoS parameters and power consumption. While the duty cycle parameter investigation shows that AA-MIMP could be enhanced but needs further research work. More investigations are also underway to extend the current work to multi-hop. We will also derive our architecture for VANET communication to demonstrate its validity and its generic character as well. REFERENCES Fig. 8: Duty Cycle vs. Flow Number notice that when the number of flows increases and the number of channels is reduced the duty cycle decreases, which means that channel s capacity is consumed. But we notice a waste of time when the number of data channels increases. This is due to unsuccessful transmissions and also to the fact that the Beacon Interval is miss-sized for current transmissions. This could be solved by a good estimation of the Beyond ATIM Window Size. H. Per bit Power Consumption Figure 9 shows better power consumption performance obtained by correct powering on used and off unused interfaces. We notice that for a low network load (number of flows less Fig. 9: Per bit Power consumption vs. Flow Number than or equal to 5), the per bit power cost is minimal for 1 data channel. For a medium network load (number of flows between 5 and 10), the energy cost of a bit is minimal for [1] The omnet++ community v-4.1, Tech. Rep. [2] J. So and N. H. Vaidya, Multi-channel mac for ad hoc networks : Handling multi-channel hidden terminals using a single transceiver, In Proc. of Mobihoc, Roppongi, Japan, [3] J. Wang, Y. Fang, and D. Wu, A power-saving multi-radio multi-channel mac protocol for wireless local area networks, In Proc. of Mobihoc, Roppongi, Japan, [4] IEEE, Ieee : Part 11 : Wireless lan medium access control mac and physical layer phy specifications, Tech. Rep. [5], Ieee b : Part 11 : Wireless lan medium access control mac and physical layer phy specifications : Higher-speed physical layer extension in the 2.4 ghz band, Tech. Rep. [6] J. So and N. H. Vaidya, Routing and channel assignment in multichannel multi-hop wireless networks with single-nic devices, Tech. Rep. [7] A. Raniwala, K. Gopalan, and T. cker Chiueh, Centralized channel assignment and routing algorithms for multi-channel wireless mesh networks, ACM Mobile Computing and Communications Review (MC2R), vol. 8, no. 2, [8] J. So, Design and evaluation of multi-channel multi-hop wireless network, Ph.D. dissertation, Urbana University, Illinois, USA, [9] R. Chandra, P. Bahl, and P. Bahl, Multinet: Connection multiple ieee networks using a single wireless card, IEEE INFOCOM, [10] A. Raniwala and T. cker Chiueh, Architecture and algorithms for an ieee based multi-channel wireless mesh network, IEEE INFOCOM, [11] IEEE, Ieee : Ieee standard for local and metropolitan area networks part 16: Air interface for broadband wireless access systems, Tech. Rep. [12], Ieee p-2010 : Ieee standard for information technology telecommunications and information exchange between systems local and metropolitan area networks specific requirements part 11: Wireless lan medium access control (mac) and physical layer (phy) specifications amendment 6: Wireless access in vehicular environments, Tech. Rep. [13], Ieee : Ieee standard for information technology local and metropolitan area networks specific requirements part 15.4: Wireless medium access control (mac) and physical layer (phy) specifications for low rate wireless personal area networks (wpans), Tech. Rep. [14] A. Dhraief, N. Montavont, J. Bonnin, and A. Belghith, Ns-2 based simulation framework to evaluate the performance of wireless distribution system, 10th Communications and Networking Simulation Symposium located with Spring Simulation Multiconference (SpringSim 07), 2007.

6 [15] R. Tahar, A. Belghith, and R. Braham, Performance evaluation of ieee multi-interface based wireless distribution system mi-wds, The 7th ACS/IEEE International Conference on Computer Systems and applications (AICCSA09), Rabat, Morocco, [16] A. Belghith, R. Tahar, and R. Braham, Enhancing qos parameters using an ieee multi-interface based wireless distribution system (miwds), The Global Information Infrastructure Symposium (IEEE-GIIS 2009), Hammamet, Tunisia, [17] S.-L. Wu, C.-Y. Lin, Y.-C. Tseng, and J.-P. Sheu, A multi-channel mac protocol with power control for multi-hop mobile ad hoc networks, THE COMPUTER JOURNAL, 2002.