Computer Engineering Bogazi¸ ci University Bebek,Istanbul 34342 Turkey
Submitted to CMPE 58C Final Project
by Melih S¨ ozdinler
1
Comparison of Routing Algorithms: Flooding, Gossiping and Rumor Routing Tuna Tu˘gcu, Melih S¨ozdinler January 11, 2010 Abstract Routing is fundamental concept of wireless networks. Since wireless concept in networks is more challenging than wired networks, it needs some specialized routing algorithms. Routing in wireless sensor networks has been accepted even more challenging traditional wireless networks. Several algorithms are proposed for routing specialized in wireless sensor networks. Due to the application specification, there is no best algorithm. In this survey paper, we compare three selected routing algorithms Flooding, Gossiping and Rumor Routing. We give several comparison cases including network size, connectivity, energy efficiency and we demonstrate prons and cons of each algorithms.
1
Introduction
Wireless Sensor Network(WSN) applications have considerable attention in the literature. Although it is studied a lot, it is still hot and goals are not achieved. WSN’s are different from other networks since the topology is very dense and Sensor Nodes have small processing units and limited batteries [3]. For this reason, trade of between cost of maintaining the model network and choosing other ways of maintenance is still stimulating. On the other hand, for some applications sensor networks are supportive. As an example, monitoring applications are widely used in harsh environments such as factory, landfills, forests and biomedical applications [7]. Furthermore, military applications of sensor networks are also useful for monitoring and protecting the area. Detailed information about sensor networks are given in a survey paper [1]. In this paper, we are considering the routing prob em of sensor networks. We still need application specific algorithms for routing. It is challenging to design 2
routing algorithms for WSN’s since energy constraint, stricted processing power constraint, living and changing environment constraints and wireless medium constraints have to be considered by routing algorithms. Each considerations lead to trade offs. Then, proposed algorithms are specialized in maximizing utility for one or more constraints rather than all constraints. In the literature, there are several proposed wireless routing algorithms. GPSR [6], Directed Diffusion [5], Rumor Routing [2], GAF [8], LEACH [4], Gossiping and Flooding. These are all specialized to maintain some goals such as obtaining data efficiency and energy efficiency. At this survey paper, in Methods section we give the definition of three selected algorithms. At the Discussion and Results section we compare Flooding, Gossiping and Rumor Routing algorithms in terms of energy efficiency, data delivery rate, average number hops and then we will discuss the prons and cons of the algorithms.
2
Methods
We compare three routing methods. First we will give detailed description of Flooding, Gossiping and Rumor Routing [2]. Showcase of Gossiping and Rumor Routing algorithms are given in Figure 1. Flooding is similar to Gossiping, since then we omit its showcase. Figure 1a and Figure 1b are sample runs of Gossiping. Furthermore, Figure 1c and Figure 1d are sample runs of Rumor Routing. Figure 1a and c are random network topology. Figure 1b and d are grid network topology. The network graph has connecting edges when two nodes are in range of each other assuming each of the nodes have constant ranges and nodes are identical.
2.1
Rumor Routing
Rumor routing is one of the data centric protocols and is also event based. When an event is created on node ne , the agent of event, ae starts to crawl over the network. During the crawl, agent tries to have almost straight path if it is possible. When a query occurs from one selected sink node nq , another agent, aq starts crawling from this sink node in a straight path. Indeed, the path, pe of an agent ae and the path, pq of an agent aq may intersect at some point. In the case of this intersection at the node ni , the path between ne to ni and ni to nq are the whole routing path from event to query. If this path exists, package delivery starts. There is no guarantee for the intersection. Both agents may stuck in area or can not leave absorbing components. Also, events and queries may have expiration header.
3
Figure 1: Routing on Random Graphs and Regular Grids (a),(b) Gossiping; (c),(d) Rumor Routing
4
2.2
Gossiping
Gossiping is one of the basic algorithm. Flooding is another version. In Gossiping, in the case of query of an specific event, source node nq selects its one of the neighbors and sends its query. Then this neighbor node also forwards the query to its selected neighbor. This procedure is applied repeatedly until a ne is gossiped. Then the whole path of the query is used for package delivery. Gossiping, indeed finds the event location, if there exist an event. However, if query has expiration header, the event acquisition may fail. In the case of stuck, nodes can give a feedback to parent nodes until a possible gossiped node can be determined by any ancestor of the stucked node. Then, normal execution continues.
2.3
Flooding
Flooding is another basic algorithm for routing. In Flooding , in the case of query of an specific event, source node nq broadcasts to the neighbors and sends its query to all of the adjacent neighbors. Then these neighbor node also forwards the query to its all adjacent neighbors that are not informed about the query. This procedure is applied repeatedly until a ne is flooded to the target event. Flooding, indeed finds the event location, if there exist an event. However, if query has expiration header, the event acquisition may fail, but since all the deliveries are in parallel respect to node it is more likely to find the target event.
3
Results and Discussions
We have several comparison methodology for both random and grid networks. Mainly, our concern is to see three routing algorithms when the network size, range, and placement changes. Due to this specialization, we made some assumptions. The first assumption is each node have constant uniform transmission range as specified before the trial and no fading occurs. The second is connections or edges between sensor nodes are stable and each sensor nodes know their positions for a given range. Moreover, when one sensor node speaks to its neighbors, consumed energy is the same for all communications. Furthermore, event creation and chosen sink is not done in parallel to simplify the implementation. At one time instance, one event and one query are running through the network. For each trial 50 event is created and they run consecutively. We limit the number of trial due to the computational complexity. This assumption makes the network traffic very low so MAC layer protocol is so simple that each nodes 5
have uniformly a package processing delay and a communication delay and sleep schedules are the same for the neighbors. In general, we assume that each sensor nodes have constant battery size and decreases one unit at each probe. Also, it decreases due to the transmission and due to the consumed processing power of package P . Finally, we have a time to leave(TTL) field for each query. If T T L = 0 for a query, we omit the corresponding packet from the network. Default T T L is 200 per query. For the implementation, we are using LEDA C++ library on a laptop computer with 2 GHZ processor and 2 GBRAM . The visualization of network windows is written with LEDA window header and plots maintained by Open Office. Experiments are in two fundamental part. First we test, the three routing algorithms by having constant range and distance between nodes while changing the size of network. Then at the second part, we change the range of all nodes in Random Networks and we change the distance between nodes in Grid Networks. The first set of experiment plots are given in Figure 4. There are four subfigures and each subfigures have two different plots. Each plots have a x coordinate as Network Size varies from 50 to 150 nodes. Furthermore, subfigures (a) and (b) show the plots for Random Networks and subfigures (c) and (d) show the plots for Grid Networks. In Figure 4a, we give two plots which are complementary of each other. According to these plots, the most efficient in terms of used energy is Gossiping. It also decreases when we increase the number of sensor nodes. The second best is Rumor Routing and it is steady state when we increase the number of sensor nodes. In Figure 4b, we show the delivery number per given number of sensor nodes. Its maximum value is 50 and achieved by Flooding. It delivers all the packets to the destination. Rumor Routing does also well. Gossiping fails to deliver the packet since its query times out due to T T L in some of the cases. It also performs worse when we increase the number of sensor nodes. At the second plot of b, Gossiping does better than others, in terms of average hops. At this plot, we omitted the queries when the target node is not found through the time of T T L. In Figure 4c and d, the corresponding plots are related to Grid Networks. The node x and y distance between neighbors are constant. In Figure 4c, the remaining energy plot of Rumor Routing is the best for sparse networks. When the network obtains some density, Flooding passes Rumor Routing. Furthermore, in Figure 4d, average hop value for Gossiping is the best for all tested network sizes. Flooding and Rumor Routing increases, their average hops with the increase in Network Size. Flooding performs best in delivery averages. Rumor Routing also approaches to the performance of Flooding while increasing the Network Size. In a conclusion, From the Figure 4, 6
Figure 2: When range changes, the topology does change due to the changes in connectivity we can make some general conclusions. Rumor Routing is not the best in all plots but it is considerable since it is not the worst. Flooding obtains fine delivery numbers but it fails for energy efficiency and average hops. Gossiping is also alternative to Rumor Routing if your applications can deal with nondelivered packets easily. In Figure 5a, we compare Energy Consumed, Average Hop and Delivery Number versus variation of distance between nodes for Grid Networks. We increase the distance d between nodes for both in x and y direction from 50 to 185 and they plotted at x axis of each plots. Rumor Routing performs well almost all plots. In general, it passes Gossiping in energy efficiency when the distance d is 95. Furthermore, after d = 140, the algorithm fails in queries due to T T L and decreases delivery number. When d increases connectivity decreases meaning less communications. Rumor Routing likes connectivity, since its decrease in delivery show us that it fails in querying process while the number of node-to-node communication is less. In Figure 5b, we compare Energy Consumed, Average Hop and Delivery Number versus variation of range for Random Networks. We show how the range of sensor nodes effects the topology in Figure 2. When range increases, the connectivity of each nodes is increased as well. At the plots in Figure 5b, Gossiping performs well at first two plots. While maintaining energy efficiency it does well in average hop number as well. When Gossiping does well in these two, interestingly it can not maintain delivery number well. This is because of its nature to choose random node to gossip. During gossip period, it tries so many neighbors so that T T L of the query becomes 0 in some cases. Due to
7
Figure 3: Random Network and Regular Grid Network
that, the sink can not able to send its packet to the target. However, it is not good for real environment such as forest fire detection. The query must reach its destination. Due to this phenomenon, Rumor Routing is again considerable since it almost deliver its queries, energy efficient and quick due to hop number. Indeed, we see that Rumor Routing and Gossiping does well in plots. Since Gossiping fails in querying the target node, it should have larger T T L values as it is used for Rumor Routing and Flooding. Flooding is not energy efficient and quick due to high average hop numbers. Rumor Routing has some problems in structured sparse Grid Networks since it tries to obtain almost straight path, sometimes it chooses long way as shown in Figure 1d.
4
Conclusion
In conclusion, we compared three wireless sensor network routing algorithm with several comparison methodologies. Briefly, Flooding has delivery guarantee with a given enough T T L. Gossiping is energy efficient but it needs a higher T T L values to obtain delivery success. Rumor Routing is the best among these algorithms. It does have steady plots, energy efficient and delivers the packet between sink and event area. Although, it is supposed to be best choice, it has some problems in structured networks due to its nature to follow straight paths. It will quickly reaches boundary nodes as we see in Grid Networks. This will also be advantageous for itself if sink nodes are always at the boundaries. Indeed, with this survey paper, we believe that we show novel comparison methodologies that will guide the future users of these algorithms.
8
Figure 4: Experiment 1 to 4: (a) and (b) are for Random Networks. (c) and (d) for Grid Networks. (a),(c) show used energy and remaining energy out of 10000; (b),(d) show delivered packet number averages(max. 50) and average hop numbers
9
Figure 5: Experiment 5 to 6: (a)Grid Networks; changing node distances, plots for used energy, average hop and average delivered packet number;(b)Random Networks; changing range parameter, plots for used energy, average hop and average delivered packet number
10
References [1] L. F. Akyildiz, W. Su, Y. Sankarasubramaniam, and E. Cayirci. A survey on sensor networks, 2002. [2] D. Braginsky and D. Estrin. Rumor routing algorthim for sensor networks. In WSNA ’02: Proceedings of the 1st ACM international workshop on Wireless sensor networks and applications, pages 22–31, New York, NY, USA, 2002. ACM. [3] D. Ganesan, D. Estrin, and J. Heidemann. Dimensions: why do we need a new data handling architecture for sensor networks? SIGCOMM Comput. Commun. Rev., 33(1):143–148, 2003. [4] W. R. Heinzelman, A. Chandrakasan, and H. Balakrishnan. Energy-efficient communication protocol for wireless microsensor networks. pages 10 pp. vol.2+, 2000. [5] C. Intanagonwiwat, R. Govindan, D. Estrin, J. Heidemann, and F. Silva. Directed diffusion for wireless sensor networking. IEEE/ACM Trans. Netw., 11(1):2–16, 2003. [6] B. Karp and H. T. Kung. Gpsr: greedy perimeter stateless routing for wireless networks. In MobiCom ’00: Proceedings of the 6th annual international conference on Mobile computing and networking, pages 243–254, New York, NY, USA, 2000. ACM. [7] L. Schwiebert, S. K. S. Gupta, and J. Weinmann. Research challenges in wireless networks of biomedical sensors. In MobiCom ’01: Proceedings of the 7th annual international conference on Mobile computing and networking, pages 151–165, New York, NY, USA, 2001. ACM Press. [8] Y. Xu, J. Heidemann, and D. Estrin. Geography-informed energy conservation for ad hoc routing. In MobiCom ’01: Proceedings of the 7th annual international conference on Mobile computing and networking, pages 70–84, New York, NY, USA, 2001. ACM.
11