You are on page 1of 6

2009 International Conference on Communications and Mobile Computing

ELQS: An Energy-efficient and Load-balanced Queue Scheduling Algorithm for Mobile Ad Hoc Networks
Jiangtao Yin, Xudong Yang Beijing Key Laboratory of Intelligent Telecommunications Software and Multimedia, Beijing University of Posts and Telecommunications, Beijing, China yinjt@buptnet.edu.cn, xdyang@bupt.edu.cn Abstract
This paper presents a novel priority queue scheduling algorithm named energy-efficient and loadbalanced queue scheduling algorithm (ELQS) for Mobile Ad Hoc NETwork (MANET) considering both mobile nodes congestion levels and the nodes energy usage, which uses mixture of energy balance and traffic balance to solve the problem of hot spots, some mobile nodes with heavy traffic load using their energy at much higher rate and die much faster than the other nodes, since mobile nodes in MANET usually operate on limited batteries. The capacity cost, a function of one mobile nodes congestion level and energy usage, is divided into three different phases by setting Min and Max two thresholds. The packets priority is dynamically assigned according to the current capacity cost of the node. The simulation results show that ELQS can effectively decrease the network transmission delay, promote the network throughput and prolong the network lifetime. network lifetime. The network performance degradation gets worse, as traffic load increases. Despite there are a large amount of effort invested in routing protocols [1, 2, 3], improving TCP performance [4, 5] and medium access control (MAC) [6, 7] for MANET, there has not been a similar focus on queue scheduling algorithms. Generally, network traffic can be classified into two categories, control packets such as routing packets that are generated by network protocols to serve for data delivery and data packets that carry payload data from the application. In a wired network, control packets are not given higher priority because of their short delay and low loss ratio. While MANET has several features, including possible frequent transmissions of control packets due to mobility, the multi-hop forwarding of packets, and the multiple roles of nodes as routers, sources, and sinks of data, so it is generally accepted that control packets should be given higher priority for scheduling. In practice, this mechanism is also implemented within a single node in most implementations. In NS-2 [8], a well-known network simulation tool, priority queueing is the default queue management policy in any mobile node. With priority queueing, these two kinds of packets are handled differently: control packets are always given higher priority than data packets in scheduling. When packets in buffer exceed the limit of the queue, data packets are always dropped first. Although it is proven to be effective and is publicly accepted, priority queueing still has several limitations. Since all control packets unconditionally are given high priority, when the amount of control packets is very large, data packets will get little or even no chance to be serviced. Also, the frequent mobility would result in the burst of control packets and then cause the nodes buffer to overflow. In addition, priority queueing does not take in account energy consumption. Since mobile nodes are energyconstrained, if a node still gives all control packets
121

1. Introduction
A Mobile Ad Hoc NETwork (MANET) is a collection of wireless mobile nodes dynamically forming a temporary network without the use of fixed network infrastructure or centralized administration, and operating on limited amount of battery energy consumed mostly in transmission and reception. MANET has known a great success and now, they are opening up to various applications having requirements of Quality of Service (QoS), such as delay, throughput, packet loss, and network lifetime. However, the mobility of nodes and the error-prone nature of the wireless medium pose many challenges, including frequent route changes and packet losses, in the way of meeting the requirements of QoS. Such challenges increase packet delay, decrease throughput and reduce
978-0-7695-3501-2/09 $25.00 2009 IEEE DOI 10.1109/CMC.2009.256

high priority when it has little energy, the note probably lies on one or more new routes and its traffic load would rise sharply. In this situation, the node will die prematurely and thereby the network may get partitioned. Therefore, in order to maximize he network lifetime, it is essential to prolong each nodes lifetime by minimizing transmission energy consumption, and sending packets via routes that can avoid nodes with low energy. This paper proposes an energy-efficient and loadbalanced queue scheduling algorithm (ELQS) to address the aforementioned issues. The capacity cost, a function of one mobile nodes congestion level and energy usage, is divided into three different phases by setting Min and Max two thresholds. The packets priority is dynamically assigned according to the current capacity cost of the node. The ELQS would avoid posing adverse impact on building routes quickly, and reduce the transmission delay, promote the network throughput, prolong the network lifetime and thereby enhance the network performance.

WR = S /(1 R )

(2)

Besides residual service time and the total service time of all the packets in queue, the waiting time of data packets includes the delay caused by new inserted routing packets. So:

T1 = RWR + DWD + S

(3)

During the waiting item of data packets, the delay caused by new inserted routing packets is: T2 = RWD (4) Thus,

WD = T1 + T2 = RWR + DWD + S + RWD (5)


Combining with (2), we have

WD = S /[(1 R )(1 R D )]

(6)

2. Overview of the priority queueing


In this section, we use queueing theory to analyze the priority queueing and then describe its limitations. We assume that the time of packets arrivals is according to the Poisson process; the service time is generally distributed; the buffer size is infinite; and routing packets are the only control packets in buffer. R denotes the arrival rate of routing packets and D represents the arrival rate of data packets. Therefore the packets arrival rate of a node is = R + D . R is the average service rate of routing packets, D is the average service rate of data packets. We define , the traffic intensity, as the average arrival rate divided by the average service rate. So R = R / R , D = D / D , the traffic intensity of the node is = R + D . S is the average residual service time, which means that when a packet is serviced, the average time the next packet need to wait to be serviced. The waiting time of routing packets in the queue is consisted of two parts: one is residual service time of the node, which means that when a packet arrives at the node if one packet is being serviced, it must wait until that packet finished its service; the other part is the total service time of all the routing packets in the queue. Thus, the average waiting time of routing packets in queue is:

From (6), we know that WD , the average waiting time of data packets in queue, is mainly influenced by the traffic intensity of routing packets that have high priority. When the number of routing packets increases, their average waiting time will not increase considerably, but the average waiting time of data packets would rise dramatically. At the same time, the priority queueing would result in quick depletion of energy of nodes along the paths if the traffic demands are long lasting and concentrated for routing protocols that are not aware of energy consumption. We define networks lifetime as the time when any node runs out of its own battery power for the first time because it can result in network partitioning and interrupt communication if a node stops its operation. The lifetime formula of the wireless ad hoc network under a given flow proposed in literature [9] is the time until the first battery drains out:

LifeTime = min
iV

ei f ij
jNi

Ei

(7)

where fij denotes the ratio between average flow on link (i,j) and the maximal possible flow on the link, N i denotes the collection of its neighboring nodes for each node i, ei denotes the transmission energy required by node i to transmit an information unit. Now the main issue of how to use energy efficiently can be expressed to maximize the lifetime of MANET, as seen in (8).

max( LifeTime) = max(min


iV

ei fij
jNi

Ei

(8)

WR = RWR + S
then we have

(1)

Unfortunately, priority queueing dose not take either the nodes buffer length occupancy or the nodes battery energy usage into consideration. Considering

122

the example in Figure 1, usage of the same shortest path would add end-to-end delay as well as load of node E (hot spot) and shorten the lifetime of the network, and hence should be avoid (the remaining energy capacity are given adjacent to the nodes.)

algorithm and we set =0.1 in the following simulations: QiAvg = (1 )QiAvg + QiCur / QLen (9) Following the idea of Battery Cost Function proposed by literature [10], Battery Cost Function is redefined for mobile node i. The Battery Cost Function f i proposed in this paper denotes the ratio of battery energy usage and total battery energy: f i = ( Ei Ri ) / E i , where Ri denotes residual battery energy capacity and E i denotes total battery energy capacity. Let ith nodes Capacity Cost Function Ci be the sum of battery cost function and normalized average queue length: Ci = f i + QiAvg = ( Ei Ri ) / E i + QiAvg (10) The larger the value of Ci of a node is, the more reluctant the node is to forward packets. We set Max and Min, two threshold values, for Ci . Then, the capacity cost is classified into three phases: low-cost phase, middle-cost phase and heavy-cost phase. We set Min=0.5, Max=0.8, = 0.8, =0.2 in the following simulations. ELQS adjust the priority of all kinds of packets according to the value of Ci . For convenience to illustrate, we select AODV as the routing protocol to describe ELQS, but ELQS does not depend on any specific routing protocol. ELQS also can accompany other routing protocols with a little modification or even without modification. In priority queueing, when the node runs out of its battery power or its buffer is full, all routing packets are still given high priority for scheduling. Therefore, in this way, a route request (RREQ) packet would get reply even when a node nearly runs out of its battery power or its buffer is full. A route reply (RREP) packet would bring a route via the node, which results in aggravating the nodes congestion and running out of its power more quickly. However, quick forwarding a route error (RERR) packet would avoid connections continuing to pass the node, and consequently reduce the nodes congestion and its energy consumption. As for ELQS, we have defined three phases according to the value of Ci . During low-cost, the nodes buffer length occupy is low and its battery energy usage is low, which means the node have enough capacity to forward all kinds of packets, so we give route packets high priority in order to build quickly a route via the node. During middle-cost, the delay of the packets and the nodes load are in balance. If we still give high priority to RREQ packets and RREP packets, one or more route via the node would be built, and then a large number of packets would be

Figure 1: Example network

3. ELQS
To eliminate the limitations of priority queueing described in section 2, in this section we propose an energy-efficient and load-balanced queue scheduling algorithm (ELQS).

Figure 2: Schematic of a mobile node The scheduler of a mobile node is positioned between the routing agent and above the MAC layer (Figure 2). The buffer queue length is an essential metric to measure a nodes traffic congestion level. Let QLen denote the length of buffer queue; QiCur and QiAvg denote the current and normalized average queue length, respectively. The average queue length QiAvg updates itself by the exponential weighted moving average (EWMA)

123

sent to the node. So we handle RREQ packets and RREP packets the way as data packets, but we give high priority to RERR packets. During high-cost, the nodes buffer length occupancy is high or its battery energy usage is high. At this time, the node have little capacity to forward packets, so we drop new arrival RREQ packets in order to avoid building new route via the node and reduce routing overhead. Since RREP packets bring important route information and their amount is small, we handle it the way as data packets, and we still give high priority to RERR packets.

each source node transmits packets at a certain rate, with a packet size of 512 bytes. We choose source and destination nodes randomly among all nodes. We vary traffic load by changing the packet sending rate in the simulations. Each point in the plotted results represents an average of five simulations run with different random mobility scenarios

4.2. Performance evaluation


We use the following performance metrics to evaluate the effect of priority queueing and ELQS: Lifetime - The time when any node runs out of its own battery power for the first time Packet delivery fraction - The ratio of the data packets delivered to the destinations to those generated by the CBR sources. Average end-to-end delay of data packets - This includes all possible delays caused by buffering during route discovery latency, queuing at the interface queue, retransmission delays at the MAC, and propagation and transfer times. Throughput - The number of data packets received by the destination node. Routing overload - The number of routing packets transmitted per data packet delivered at the destination. Each hop-wise transmission of a routing packet is counted as one transmission.

4. Performance evaluation
In this section we describe our simulation environment and performance metrics, and then illustrate simulations results.

4.1. Simulation Environment


We have implemented our algorithm in NS-2 [8]. NS-2 includes the simulation model for mobile ad hoc networks developed by the CMU Monarch project. The model includes a physical layer, an 802.11 MAC layer, and a data link layer [3]. Its radio model is very similar to the Lucent WaveLAN radios with nominal radio range of 250m, and the nominal bit rate is 2 Mbps. As mentioned earlier, we performed our algorithm with AODV as the routing protocols. We used 100 wireless mobile nodes, moving in a rectangular grid of dimensions 2200m x 600m to form a MANET. Each simulation was executed for 300 simulated seconds. Nodes in simulations moved according to the random waypoint model [3]. We controlled the degree of mobility through the pause time. We used pause times of 20 seconds and the maximum speed for a node is 20 meters per second. The energy model in NS-2 is used to measure energy consumption. It has three states where energy is consumed: transmitting, receiving and idle state. Every node in NS-2 starts with initial value which is the amount of energy defined by user at the beginning of the simulation. The nodes physical layer also requires transmitting power (TXpower), receiving power (RXpower) and idle power parameters, and these values also can be defined by user. Initial energy is decremented for transmission and reception of packets by TXpower and RXpower. When energy in a node becomes zero, the node does not accept or send any packets. We set the TX power to 0.6W, RX power to 0.3W, Idle power to 0.1W, and set initial energy to 50J in our simulations. We used a constant bit rate (CBR) source as the data source for each node. There are 10 sources, and

4.3. Simulations results and analysis


We present in this subsection the performance of priority queueing and ELQS for the various metrics presented above. We vary the packet sending rate to reflect various traffic loads.

Figure 3: Network lifetime versus packet sending rate

124

The simulation results in Figure 3 show that ELQS exceeds Priority Queueing in network lifetime especially when much heavier traffic (larger packet sending rate).

Figure 6: Packet delivery fraction versus packet sending rate Figure 6 shows the packet delivery fraction difference between ELQS and Priority Queueing. The percent of data packets delivering decreases as the packet sending rate increase because of congestion, but the packet delivery fraction is higher in ELQS than Priority Queueing when packet sending rate becomes large.

Figure 4: Average end-to-end delay versus packet sending rate The simulation results in Figure 4 and 5 showed differences of average end-to-end delay and throughput between Priority Queueing and ELQS. It is evident that with ELQS the network performance, mainly including average delay and throughput, is much better than that of Priority Queueing, because ELQS tries to relieve mobile node with heavy load, expressed by large buffer occupancy and large usage ration of battery energy and make traffic more balanced. As a result delay is reduced, the lifetime of the MANET is prolonged and network throughput is increased.

Figure 7: Routing overhead versus packet sending rate Figure 7 shows the routing overhead comparison between ELQS and Priority Queueing. It indicates that the routing overhead increases as the packet sending rate increase due to increase in number of route requests and number of route replies transmitted in the network. ELQS generates less routing overhead compared to Priority Queueing, since it would drop route request or not give it high priority when traffic load is heavy or energy usage is high.

Figure 5: Throughput versus packet sending rate

125

From all the above-mentioned results, it can be concluded that ELQS has better performance than Priority Queueing.

5. Conclusion
In this paper, a novel energy-efficient and loadbalanced queue scheduling algorithm for MANET is presented and stimulated. The capacity cost, a function of one mobile nodes congestion level and energy usage, is divided into three different phases by setting Min and Max two thresholds. The packets priority is dynamically assigned according to the current capacity cost of the node. The simulation results obviously indicate that much better traffic balance effect and maximum network lifetime could be achieved with the presented mechanism. Furthermore, our scheme also performs better than Priority Queueing in terms of packet delivery fraction and routing overhead. In future, we would like to study more routing protocols and different data sources. Studying more routing protocols will help us to see the broader effect of the ELQS algorithm. We use CBR sources in this paper, but studying different data sources such as bursty TCP sources will help us to understand how the ELQS algorithm affect network performance with these sources.

6. Acknowledgements
This work is supported by Beijing Municipal Commission of Education (Con-constructing Program No. SYS100130422).

[1] Charles E. Perkins, Elizabeth M. Royer, and Samir Das, Ad Hoc On Demand Distance Vector (AODV) Routing, IETF Internet draft, Mobile Ad-hoc Network Working Group, IETF, January 2002. [2] David B. Johnson, David A. Maltz, Yih-Chun Hu, and Jorjeta G. Jetcheva, The Dynamic Source Routing Protocol for Mobile Ad Hoc Networks, IETF Internet draft, Mobile Ad-hoc Network Working Group, IETF, February 2002. [3] J. Broch, D.A. Maltz, D.B. Johnson, Y.C. Hu, and J. Jetcheva, A performance comparison of multi-hop wireless ad hoc network routing protocols, In Proceedings of ACM/IEEE MOBICOM, Dallas, USA, October 1998, pp.8597 [4] K Chandran, S Raghunathan, S Venkatesan Ravi Prakash, A feedback- based scheme for improving TCP performance in ad hoc wireless networks, IEEE Personal Communications Magazine, 2001, 8(1) pp. 34-39 [5] J Liu, S Singh, ATCP: TCP for mobile ad hoc networks, IEEE Journal on Selected Areas in Communications, 2001, 19(7), pp.1300-1315 [6] H. Luo, S. Lu, and V. Bharghavan, A New Model for Packet Scheduling in Multihop Wireless Networks, In Proceedings of ACM MOBICOM, Boston, USA, August 2000, pp.76-86 [7] V. Kanodia, C. Li, A. Sabharwal, B. Sadeghi, and E. Knightly, Ordered Packet Scheduling in Wireless Ad Hoc Networks: Mechanisms and Performance Analysis, In Proceedings of ACM MOBIHOC, Lausanne, Switzerland, July 2002, pp.58-70 [8] NS-2, Network Simulator, Version 2, home page: http://www.isi.edu/nsnam/ns/ [9] Jae-Hwan Chang Tassiulas, L., Energy conserving routing in wireless ad-hoc networks, In Proceedings of IEEE INFOCOM, Tel Aviv, Israel, March 2000, pp. 22-31

7. References

[10] Tohin, C. K., Maximum battery life routing to support ubiquitous mobile computing in wireless ad hoc networks, IEEE Communications Magazine, 39(6), pp.138-147

126

You might also like