Performance Evaluation of Failed Link Detection in Mobile Ad Hoc Networks

Transcription

1 Performance Evaluation of Failed Link Detection in Mobile Ad Hoc Networks Dimitri Marandin Chair for Telecommunications, Electrical Engineering Department Technical University of Dresden, Dresden, Germany Abstract Many ad hoc routing protocols need to determine link failures in the network. For that they keep track of the "next hop" nodes, which are used as next relay points in the routes. Hence the node must be able to recognize the case when a link to a next hop is failed. Most ad hoc routing protocols may detect broken links using hello messages or feedback provided to the protocol by the Medium Access Control(MAC) layer. While MAC feedback works better than hello messages at low network load we identified that if the traffic load on the network is high, the amount of incorrect link failure decisions from the MAC layer feedback also dramatically increases resulting in lower throughput. In this paper we address advantages and drawbacks of both approaches and investigated the delay of MAC feedback approach. I. INTRODUCTION Wireless ad hoc networks [1,2,3] are used where a stationary communication infrastructure is lacking and is expensive or infeasible to deploy (e.g. disaster relief efforts, battlefield, etc.). The military applications are still the main applications of ad-hoc networks, although there is a trend to apply ad hoc networks also for commercial uses because of their unique properties. Due to power limitation each station has a fixed range. It also acts as a router, relaying data packets for other stations to their final destination. One of the main challenges in the design of these networks is the routing mechanism. [4] upon a dynamically changing topology, node power constraints and the properties of the wireless channel. In recent years wireless ad hoc radio network have received a tremendous amount of attention from researchers. 398 IEEE WLAN[5] with access points became omnipresent. These networks are well suited for a variety of traffic types, including multimedia. One of major problem of infrastructure-based networks is the complex deployment and configuration of these networks. Ad hoc networking is not subject to these constraints. Numerous ad hoc routing protocols need to determine link failures in the network. In this paper an implementation of the Ad hoc On-Demand Distance Vector (AODV) routing protocol [6] is utilized to determine the effectiveness of link failure detection. The AODV Routing protocol is used as a representative of on-demand routing protocols meaning that it builds routes between nodes only as desired by source nodes. AODV keeps a routing table on each station in the ad hoc network. When a station tries to transmit a data packet to a destination node for which the route is not known, it uses a route request / route reply query cycle to find a route. Route discovery works by flooding the network with route request (RREQ) packets. At the time RREQ is forwarded routing table entries (reverse path entries) are created pointing back to the source. Each node receiving a RREQ, re-broadcasts it, unless it is either the destination or it has a route to the destination. Such a node unicasts the RREQ with a route reply (RREP) packet that is routed back to the original source. Routing table entries created at the time RREQ was forwarded are used to route the RREP to the source. The reverse path entries are expired after a short period of time. As the RREP propagates back to the source, nodes create forward path entries pointing to the destination. Once the source node receives the RREP, it may begin to send data packets to the destination. If any link on a route is failed, the source will be informed by a route error (RERR) packet. The source

2 and any intermediate node on the route of the RERR packet delete the broken route from their routing tables. The RERR propagation works in the following manner. A set of predecessor nodes is kept for each routing entry on every node. It points to the set of neighbouring nodes that use that routing entry to route packets. These nodes are informed with RERR packets when the next hop link breaks. Each predecessor node forwards the RERR to its own predecessors, so effectively removing all routes using the failed link. A routing table entry is expired if not used recently. Only useful routes are maintained to keep routing overheads low. AODV also uses a sequence number-method to determine actuality of routing table entries. It also help to guarantee loopfree work of the protocol. More details of AODV can be found in [6]. Many ad hoc routing protocols (for example, AODV, one of the most popular) keep track of the next hop nodes that are used as next relay points in the routes. Hence the node must be able to recognize the case when a link to a next hop is failed. Routing effectiveness and therefore link failure detection is key problem in mobile ad hoc networks. The rest of the paper is organized as follows. In section 2 two approaches of failed link detection in ad hoc routing are presented: messages and MAC layer feedback. To gain a better insight into MAC layer feedback section 3 gives the short overview of IEEE MAC DCF. Section 4 presents a performance evaluation and the discussion of simulation results when comparing MAC layer feedback and hello messaging. Finally section 5 concludes the paper and outlines some questions for further research. II. DETECTION OF FAILED LINKS IN AD HOC NETWORKS Ad hoc routing protocols may detect broken links using 1) hello messages, 2) feedback provided to the protocol by the MAC layer and 3) passive acknowledgements 1) messages The reason of using hello messages to determine link existence come from the assumption that receiving of a hello message signifies link availability with the source of the hello. This method works well on wired networks, which suffer from few packet losses and topology changes. In order to keep up routes, AODV usually demands that each node transmits a hello message at regular intervals (if the node has not broadcasted any other control messages during the previous second), with a default rate of e.g. once per second. Inability to receive three successive hello messages from a neighbour is interpreted as a sign that the link to the given neighbour is failed. When AODV is run over IEEE , messages do not need to be used due to the MAC layer feedback of unreachable next hops. When combined with the other MAC protocols, however, messages are needed since such feedback is not available. Many current implementations of routing protocols rely on hello messages. 2) MAC Feedback Alternatively, the AODV standard proposes that a station may use MAC layer methods to find out link failures to neighbouring nodes. This approach gives the routing protocol the possibility to quickly find broken links. MAC layer feedback are callbacks to the network layer sent by the MAC layer explicitly declaring a transmission error indicating that a packet could not be forwarded to its next hop node. 3) passive acknowledgements[11] If MAC layer feedback is not available, DSR specifies other approach, known as passive acknowledgments, in which a node, after a packet transmission to the next hop on the route, continues to listen the channel and overhears whether the next hop forwards the packet further along the path. If it doesn t hear the forwarding of the packet during predefined time, it draws a conclusion about link failure. The mechanism of passive acknowledgments suffers from the fact that it requires from WLAN network cards a support of promiscuous mode, which is extremely energy-expensive. Ericsson Simulation Work[10] showed that a low power devices such as Bluetooth consume roughly 50% more energy as the receiver would frequently need to decode all packets besides its own packets. As passive acknowledgements didn t find wide application in ad hoc networks research from aforementioned reasons, we restrict our focus in this paper on most used approaches: hello messages and MAC feedback. 399

3 Fig 1. MAC Feedback delay In this paper we addressed advantages and drawbacks of both approaches. We also investigated the delay of MAC feedback approach. Although large number of papers have performed simulations using one of the mechanisms, no papers, as far as we know, clearly compare them in terms of detection delay, energy consumption, impact on network throughput. [7] examined the effectiveness of hello messaging for monitoring link status. In this study, it is concluded that many parameters influence the utility of hello messages. The authors examine these factors and experimentally evaluate a variety of approaches for improving the accuracy of hello messages as an indicator of local connectivity. [8] simulated ad hoc routing protocols when run over different MAC protocols. It was determined that the choice of MAC layer protocol affects the relative performance of the routing protocols. Because some routing protocols require periodic hello messaging when run over link layer protocols that do not provide MAC feedback when the next hop is unreachable, the amount of control traffic generated with other MAC protocols is considerably greater than with IEEE DCF III. IEEE MAC MAC FEEDBACK In a typical ad hoc wireless network mobile stations use omni-directional antennas, so the channel is to be shared by nearby situated stations. The sharing of this channel is managed by the IEEE Medium Access Control (MAC) protocol. The IEEE MAC layer may operate in two different channel access mechanisms: the point coordination function (PCF) and distributed coordination function (DCF). The distributed coordination function (DCF) of IEEE became practically the standard MAC protocol for ad hoc networking and is now in use in almost all simulations and testbeds for wireless multihop ad hoc network investigation. As the PCF cannot be utilized in ad hoc networks further we consider only the DCF. The DCF is in essence a carrier sense multiple access with collision avoidance (CSMA/CA) and includes the use of RTS/CTS mechanism. The CSMA/CA protocol manages access to the wireless medium, to avoid messages getting lost due to frame collisions. This method is actually inherited from Ethernet (CSMA/CD) and is based on packet contention. A CSMA/CA referred to as the Physical Carrier sense operates as follows: When a node receives a packet to be sent from the higher layer, it listens from the beginning to ensure that no other nodes are transmitting. If the channel is idle for a period of time equal to a distributed interframe space (DIFS), then the station may transmit the packet. Else, it defers its transmission by selecting a random "backoff" period which determines the interval of time the node must wait until it is permitted to attempt to retransmit its packet. During periods in which the channel is sensed as idle, the transmitting node decrements its backoff counter. When the channel is busy, backoff counter is frozen. When the backoff counter runs down to zero, the node is allowed to retransmit the packet. As the probability that two nodes will select the same backoff period is small, collisions between packets are minimized. But there is always a possibility that some stations can sense channel as idle at once and transmit at the same time, invoking a collision. Collision detection, as is used in Ethernet, cannot be employed for the radio transmissions of IEEE The cause for this is that when a node is transmitting it cannot hear any other nodes in the network which may be transmitting, since its own signal will suppress any others reaching at the node. To prevail over the collisions (or at least to perform a fast collision detection) and resolve well known hidden terminal problem, which occurs 400

4 Goodput Energy efficiency Overall recieved packets MAC feedback Energy, Joules 0,08 0,07 0,06 0,05 0,04 0,03 0,02 0,01 0 MAC Feedback Fig. 2. Goodput with different link failure detection approaches for static network Fig. 3. Energy consumption per received packet with different link failure detection approaches for static network often in ad hoc networks the RTS/CTS dialogue was developed. Before actual data transmission can take place, RTS and CTS handshaking packets are used to reserve the channel. The station, that has a data packet to send, first transmits a Request To Send (RTS) control frame to the intended receiver of the data packet. If the intended receiver hears the RTS, it responds with a short clear-to-send (CTS) packet after a period of time called Short Interframe Space(SIFS). Any other stations hearing either of both packets must be silent and defer transmission by setting Network Allocation Vector (NAV) an internal structure that records when the channel might be free. Using the NAV a station knows when the current transmission ends. The use of NAV is referred to as the Virtual Carrier sense. The transmitting station is allowed to transmit its data packet only if the CTS frame is properly received. Since the CSMA/CA can not detect a collision, the acknowledgment ACK for data packet is transmitted after SIFS interval. After 7 (according to IEEE ) RTS retransmissions have failed MAC sends the routing protocol a message that the link is failed. So the delay of MAC feedback to detect link failure consists of following parts: - waiting a random backoff time after the last packet transmission, as the node must wait between two consecutive transmission - DIFS period before sending RTS - DIFS period for waiting CTS - backoff time between RTS retransmissions As the backoff timer is frozen if the channel is busy, backoff waiting time can be large in networks with high load. IV. EEE MAC MAC FEEDBACK PERFORMANCE EVALUATION We did several simulations using Network Simulator (NS-2)[9] for different link failure detection approaches (hello messages, MAC feedback). The radio interface of nodes was based on Lucent s WaveLan with 250m of propagation range. Nodes used omni-directional antennas. We used the following network configuration. The network size was fixed to 50 nodes located in a 670mx670m square. They use a two-ray ground propagation model and a random-way point model as the mobility model. In this model, nodes choose a random destination point and a speed value from uniform distribution between 0 and 20m/s. When they reach their destination, the mobility model is repeated. The IEEE MAC protocol with Distributed Coordination Function (DCF) is used as the MAC layer in the experiments. A traffic load was generated to simulate constant bit rate sources. 5, 10, 15, 20, 25, 30, 35 data sessions with randomly selected sources and destinations are simulated. Each node transmits data packets at a rate of four packets/sec. The size of data payload was 512 bytes.. The total simulation run time lasted 900 seconds with a warm up period

5 Goodput Energy efficiency Overall received packets MAC Feedback Energy, Joules 0,07 0,068 0,066 0,064 0,062 0,06 0,058 0,056 0,054 0,052 MAC Feedback Fig. 4. Goodput with different link failure detection approaches for mobile network seconds. All communication flows started during first 200 seconds. Results presented are an average of 10 runs (excluding warm-up period). We concentrated on two performance metrics in simulation: goodput and energy efficiency. Goodput denotes the number of packets that is correctly received. It is in essence proportional to throughput. The energy efficiency of packet transmissions refers to the ratio of the total consumed energy of all nodes to the total number of successfully received packets at the destinations. It is often essential to optimise energy expenditure in a mobile ad hoc network. Last one being to estimate overhead cost in the network. Simulations involving the measurement of goodput of the connections gave the following results about the relative performance achieved with hello messages and MAC layer feedback. As shown in figure 2 both detection methods provide similar goodput if the number of CBR connections is low or moderate in the static ad hoc network. But if the traffic load in the network increases, hello messages predominate MAC feedback as the number of incorrect link failure decisions of MAC Feedback increases. The average of energy expenditure per received packet has an opposite trend (see figure 3), as 1) with moderate traffic the MAC feedback approach detects link failure very well without energy consumption related to broadcasting of hello messages 2) with heavy traffic load an energy consumption for overhead burden increases due to many incorrect link failure detection decisions. In mobile networks MAC feedback shows better performance than the hello messages (see figure 4). Fig. 5. Energy consumption per received packet with different link failure detection approaches for mobile network But if the network load grows its goodput decreases then energy consumed per received packet (overhead) increases (see figure 5). It is interesting to note from comparison of figures 2 and 4 that in term of network goodput, hello messages are optimal for a static network, while the MAC feedback is better for a mobile network. MAC layer has to perform multiple RTS transmissions before it can conclude anything about the link failure. It results in significant delay between the occurrence of link failure and its determination (see figures 6 and 7). This detection time increases with increasing load in the network as 1) the number of retransmissions increases 2) the backoff timer is frozen for long time. If the load is low or moderate then MAC Feedback provides quick link failure detection, where as periodic hello messages at the routing layer are more quick and robust in overloaded networks. We investigated the delay of route failure detection with MAC feedback and we found out that it is normally less than delay with hello messages (3 seconds) but in some rare cases it can achieve 9 seconds (see figure 6) in mobile networks with high load. Figure 7 shows the delay in dependence on the sum of number of active neighbour stations and the number of active hidden stations relative transmitter of RTS. An active neighbour station is a station which is located in the range of the transmitter of RTS and has a packet to transmit. An active hidden station is a station in the range of intended receiver of RTS which is not in the range of transmitter of RTS and has a packet to transmit. The number of active 402

6 pdf Fig. 6. Complimentary CDF of delay of route failure detection with MAC feedback from number of CBR connections hidden stations has also an impact on MAC feedback delay. If this number is large than the probability increases that the receiver of RTS can not answer with CTS due to virtual carrier sense. It results in a setting of backoff timer on the transmitter of RTS and a necessity of new retransmission. Both will increase the MAC feedback delay. For example, the delay is significantly larger with 15 active neighbouring and hidden nodes in comparison with 2 or 5 (see figure 7). V. CONCLUSIONS Ad hoc routing protocols may detect broken links using hello messages, feedback provided to the protocol by the MAC layer or passive acknowledgements. As passive acknowledgements didn t find wide application in ad hoc networks research, we addressed advantages and drawbacks of first two approaches in this paper. While MAC feedback works better than hello messages with small network load we identified that if the traffic load on the network is high, the amount of incorrect decisions about link failures that the MAC layer make also dramatically increases that results in lower throughput. The revealing link failures based on MAC layer feedback may not be reliable if the traffic in the network increases. The reason for this behaviour is that IEEE DCF, used in our simulation, increases the amount of RTS collisions as the traffic load increases. The collisions are so frequent that the MAC layer (after 7 RTS retransmission according to IEEE ) sends incorrect feedback to the routing protocol, a sign that initiates AODV to send a Route 403 Fig. 7. Delay of route failure detection with MAC feedback from sum of active neighbours and hidden terminals Error to the source. In that way the source node then floods a network with Route Request control packet. As the traffic load on the network grows, the amount of incorrect decisions about link failures that the MAC layer reports also dramatically increases. This increments the amount of Route Requests being broadcasted on the network and induces additional overhead. We have discovered that most of the collisions, occurred due to virtual carrier sense, happen in this scenario: the sender transmits RTS to next hop neighbour, but it could not answer with CTS as its NAV was set. So many incorrect decisions about link failures are made as the intended receiver of RTS couldn t transmit CTS back due to NAV, but in the reality the link exists. The MAC feedback method reduces the overhead of the regular hello messages (figures 5 and 7), but partly modifies the protocol; for instance, all link failure detection in AODV is solely on-demand, and from that a failed link cannot be recognized until a packet has to be sent over the link. In contrary, using the periodic hello messages in standard AODV allows failed links to be detected before a packet is sent to the next hop. But as AODV needs periodic broadcasting when run over MAC layer protocols that do not provide feedback about the presence of the next hop, the consumed energy and the volume of control traffic caused with these MAC protocols are significantly greater than when it is run over MAC IEEE

7 REFERENCES [1] I. Chlamatac, M. Conti, and J. Liu "Mobile Ad Hoc Networking: Imperatives and Challenges"; Ad Hoc Networks 1 (2003) [2] Borriello G, "Key challenges in communication for ubiquitous computing"; IEEE Communications Magazine 40(5):16-18, [3] R.Ramanathan, J. Redi "A brief overview of ad hoc networks - challenges and directions", IEEE Communications Magazine 40(5):20-22, [4] John Novatnack, Lloyd Greenwald and Harpreet Arora "Impact of ad hoc routing protocols on pervasive computing" Under review for IEEE/PerCom 2004, Second IEEE International Conference on Pervasive Computing and Communications, Orlando, Florida, March [5] IEEE Standard for Wireless LAN Medium Access Control(MAC) and Physical Layer(PHY) Specifications, P [6] C. E. Perkins and E. M. Royer, Ad-Hoc On-Demand Distance Vector Routing, Proceeding of the IEEE workshop on Mobile Computing Systems and Applications, page , February available from Oct 1996 [7] Ian D. Chakeres and Elizabeth M. Belding-Royer. "The Utility of Messages for Determining Link Connectivity." Proceedings of the 5th International Symposium on Wireless Personal Multimedia Communications (WPMC) 2002, Honolulu, Hawaii, October 2002, pp [8] Elizabeth M. Royer, Sung-Ju Lee and Charles E. Perkins. "The Effects of MAC Protocols on Ad hoc Network Communications." Proceedings of the IEEE Wireless Communications and Networking Conference, Chicago, IL, September [9] "ns-2 Network Simulator", Obtain via [10] Report on 44th IETF Meeting in Minneapolis, Minnesota. h-99mar-ietf-92.html [11] John Jubin and Janet D. Tornow. The DARPA Packet Radio Network Protocols. Proceedings of the IEEE, 75(1):21-32, January